[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-tao-train-foundation-stereo":3,"mdc--r0o6vm-key":34,"related-org-nvidia-tao-train-foundation-stereo":2231,"related-repo-nvidia-tao-train-foundation-stereo":2390},{"slug":4,"name":4,"fn":5,"description":6,"org":7,"tags":11,"stars":23,"repoUrl":24,"updatedAt":25,"license":26,"forks":27,"topics":28,"repo":29,"sourceUrl":32,"mdContent":33},"tao-train-foundation-stereo","train FoundationStereo depth estimation models","Stereo depth estimation using FoundationStereo. Predicts disparity maps from stereo image pairs for 3D reconstruction. Use when training, evaluating, exporting, or running inference for a TAO FoundationStereo model. Trigger phrases include \"train stereo depth\", \"FoundationStereo\", \"stereo disparity estimation\", \"3D reconstruction from stereo\".",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},"nvidia","NVIDIA","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fnvidia.png",[12,16,17,20],{"name":13,"slug":14,"type":15},"Deep Learning","deep-learning","tag",{"name":9,"slug":8,"type":15},{"name":18,"slug":19,"type":15},"Imaging","imaging",{"name":21,"slug":22,"type":15},"Computer Vision","computer-vision",2473,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills","2026-07-14T05:27:05.109297","Apache-2.0",281,[],{"repoUrl":24,"stars":23,"forks":27,"topics":30,"description":31},[],"AI agent skills published by NVIDIA","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills\u002Ftree\u002FHEAD\u002Fskills\u002Ftao-train-foundation-stereo","---\nname: tao-train-foundation-stereo\ndescription: Stereo depth estimation using FoundationStereo. Predicts disparity maps from stereo image pairs for 3D\n  reconstruction. Use when training, evaluating, exporting, or running inference for a TAO FoundationStereo model. Trigger\n  phrases include \"train stereo depth\", \"FoundationStereo\", \"stereo disparity estimation\", \"3D reconstruction from stereo\".\nlicense: Apache-2.0\ncompatibility: Requires docker + nvidia-container-toolkit.\nmetadata:\n  version: \"0.1.0\"\n  author: NVIDIA Corporation\nallowed-tools: Read Bash\ntags:\n- stereo\n- depth\n- estimation\n---\n\n# Depth Net Stereo\n\nStereo depth estimation using FoundationStereo architecture. Predicts disparity maps from stereo image pairs for 3D reconstruction.\n\nUses pretrained Depth Anything v2 and EdgeNeXt encoders. Set `model.stereo_backbone.depth_anything_v2_pretrained_path` and `model.stereo_backbone.edgenext_pretrained_path`.\n\nThe mono and stereo skills both invoke the unified TAO `depth_net` CLI inside the container; the mono\u002Fstereo family is selected via `model.model_type` (e.g., `FoundationStereo`).\n\nPyT actions packaged by this model skill: `train`, `evaluate`, `inference`, `export`, and `quantize`. The PyT `depth_net` entrypoint does not accept a `gen_trt_engine` action in the current TAO image; build TensorRT engines only through the deploy workflow.\n\nFor TAO Deploy TensorRT actions (`gen_trt_engine`, TensorRT `evaluate`, and TensorRT `inference`), read `references\u002Ftao-deploy-foundation-stereo.md` first. The deploy spec template lives in this skill's `references\u002Fspec_template_deploy.yaml`.\n\n## Train Action Policy\n\nThis model is AutoML-enabled at the model layer. Before handling any train-stage request, read `references\u002Fskill_info.yaml` and resolve the run override from either an explicit `automl_policy` value or the user's workflow request. Use `automl_policy: on` by default and only expose `on` \u002F `off` in new launch prompts. Treat phrases like \"turn off AutoML\", \"disable AutoML\", \"no HPO\", or \"plain training\" as `automl_policy: off` for this run only. When `automl_policy: on`, `automl_enabled: true`, and both `schemas\u002Ftrain.schema.json` and `references\u002Fspec_template_train.yaml` are packaged, route the train action through `tao-skill-bank:tao-run-automl` by default with this model's `skill_dir`. Preserve workflow\u002Fapplication overrides for datasets, specs, output directories, GPU\u002Fplatform settings, parent checkpoints, and `automl_policy`. Use direct model training only when `automl_policy: off` or the packaged train schema\u002Ftemplate is missing; in the missing-schema case, report that AutoML is enabled but not runnable for this model until schemas are generated.\n\nNon-train actions such as `evaluate`, `inference`, `export`, and deploy flows stay in this model skill. The per-run `automl_policy` override does not change model metadata.\n\n## Workflow\n\n### Prerequisites — data accessibility\n\nYour dataset (left + right images + GT disparity) must be reachable from inside the container:\n- **SDK runner**: place files at the S3 paths the runner resolves (the `S3_TRAIN` \u002F `S3_EVAL` placeholders shown in the spec overrides). The runner handles S3 → container-path mounting transparently.\n- **Direct `docker run`** (e.g. local testing): mount the host dataset root read-only at the same in-container path:\n\n```\ndocker run ... -v \u003Chost_data_root>:\u003Chost_data_root>:ro \u003Ccontainer> ...\n```\n\nThe same accessibility requirement applies to the `\u003Coutput_dir>` written by all actions.\n\n### Step 1 — Annotation file\n\nPer-line annotation file referenced by `data_sources[*].data_file`:\n\n| Columns | Format | Use |\n|---|---|---|\n| 2 | `\u003Cleft> \u003Cright>` | Stereo inference (no GT) |\n| 3 | `\u003Cleft> \u003Cright> \u003Cdisparity>` | Stereo with GT |\n| 4 | `\u003Cleft> \u003Cright> \u003Cdisparity> \u003Cocclusion_mask>` | Stereo with GT and occlusion mask |\n\nIf you already have one, point to it. Otherwise generate via `depth_net convert`:\n\n```\ndepth_net convert -e \u003Cconvert_spec.yaml>\n```\n\n`convert_spec.yaml` template (stereo):\n\n```yaml\nresults_dir: \u003Cdirectory where generated annotation files are written>\ndata_root: \u003Cdirectory whose immediate children are scene folders that contain your image+depth files; convert walks data_root recursively but expects per-scene subdirectories at one level below>\nimage_dir_pattern: [\u003Csubstring matching left image paths>]\nright_dir_pattern: [\u003Csubstring matching right image paths>]\ndepth_dir_pattern: [\u003Csubstring matching GT disparity paths>]\nnocc_dir_pattern: []                 # optional, occlusion mask paths\nimage_extension: '.png'  # always include the leading dot\ndepth_extension: '.png'  # form must match image_extension (the swap is a substring replace)\nnocc_extension: ''\nsplit_ratio: 0.0        # 0.0\u002F1.0 = test-only; 0.8 = 80\u002F20 train+val\n```\n\n`convert` walks `data_root` recursively, selects paths whose path-string contains *all* substrings in `image_dir_pattern` (AND-filter), then derives right \u002F depth \u002F mask paths by replacing `image_dir_pattern[0]` with the corresponding pattern's first element plus extension swap. Inspect your dataset's directory layout and identify the substrings distinguishing left, right, and GT (e.g. `im0` vs `im1` vs `disp0GT` for Middlebury).\n\n### Step 2 — Pair `model_type` and `dataset_name` based on your data\n\nPrefer the dataset-specific class when your layout matches a supported one — it applies class-specific path conventions, evaluation crops, and (where applicable) occlusion-mask handling. Fall back to `GenericDataset` only for layouts that do not match any registered class.\n\n| Data category | `model_type` | `dataset_name` |\n|---|---|---|\n| Middlebury data | `FoundationStereo` | `Middlebury` |\n| KITTI data | `FoundationStereo` | `Kitti` |\n| ETH3D data | `FoundationStereo` | `Eth3d` |\n| FSD synthetic data | `FoundationStereo` | `FSD` |\n| IsaacReal synthetic data | `FoundationStereo` | `IsaacRealDataset` |\n| Crestereo synthetic data | `FoundationStereo` | `Crestereo` |\n| Other \u002F non-canonical layout | `FoundationStereo` | `GenericDataset` |\n\nValid `dataset_name` values for stereo `data_sources` (case-insensitive): `FSD`, `IsaacRealDataset`, `Crestereo`, `Middlebury`, `Eth3d`, `Kitti`, `GenericDataset`.\n\nThe same `dataset_name` value applies across train and evaluate actions (all of which use 3-column or 4-column annotations with GT disparity). The deploy-side `evaluate` action follows the same rule — see `references\u002Ftao-deploy-foundation-stereo.md`. For inference with 2-column annotations (left + right, no GT), use `dataset_name: GenericDataset` regardless of data layout — the dataset-specific classes (`Middlebury` \u002F `Kitti` \u002F `Eth3d` \u002F `FSD` \u002F `IsaacRealDataset` \u002F `Crestereo`) require 3-column input and reject 2-column annotations at the dataloader level. For inference with 3-column annotations (left + right + GT), the dataset-specific class is fine.\n\n### Step 3 — Write spec yaml from the spec overrides\n\nCopy the action block from `references\u002Fspec-overrides-foundation-stereo.md`. Replace:\n- `model.model_type` from Step 2 (typically `FoundationStereo`)\n- `dataset.\u003C...>.data_sources[*].dataset_name` from Step 2\n- `dataset.\u003C...>.data_sources[*].data_file` with the path from Step 1\n- For deploy-side `evaluate`: enforce `dataset.test_dataset.batch_size: 1` (see `references\u002Ftao-deploy-foundation-stereo.md`).\n\nShape consistency: the `crop_size` in `dataset.test_dataset.augmentation.crop_size` should match `export.input_height` \u002F `input_width` so the trained-model evaluator and the deploy-side TensorRT evaluator operate at the same shape. Note that `crop_size` is decorative on the pyt `evaluate` path but authoritative on the deploy `evaluate` side — see `references\u002Ftroubleshooting-foundation-stereo.md` and `references\u002Ftao-deploy-foundation-stereo.md`.\n\nFresh-install smoke runs are validated at `crop_size: [128, 128]` with `dataset.max_disparity: 128` and `model.max_disparity: 128`. Avoid 112×112 crops and avoid setting `max_disparity` smaller than the square crop side for smoke tests: those combinations can fail inside FoundationStereo with feature-map or loss-mask shape mismatches before a checkpoint is produced.\n\nData source overrides are **mandatory for every action**. Each `data_sources` entry is a dict with two mandatory fields: `data_file` and `dataset_name`. See `references\u002Fspec-overrides-foundation-stereo.md` for the per-action dataset-requirements table, every action's override block, and the `quantize` known-issue note.\n\n### Step 4 — Run\n\nCreate writable home\u002Fcache directories inside the mounted output path before using\n`--user`. Some TAO containers do not have an `\u002Fetc\u002Fpasswd` entry for the host UID,\nand PyTorch \u002F matplotlib need writable cache paths when running as that UID.\n\n```bash\nmkdir -p \u003Coutput_dir>\u002Fhome \\\n         \u003Coutput_dir>\u002F.cache\u002Fmatplotlib \\\n         \u003Coutput_dir>\u002F.cache\u002Ftorchinductor \\\n         \u003Coutput_dir>\u002F.cache\u002Fxdg\n```\n\n```\ndocker run --gpus 'device=0' --shm-size 16G --ipc=host \\\n  --user \"$(id -u):$(id -g)\" \\\n  -e USER=\"$(id -un)\" \\\n  -e LOGNAME=\"$(id -un)\" \\\n  -e HOME=\u003Coutput_dir>\u002Fhome \\\n  -e MPLCONFIGDIR=\u003Coutput_dir>\u002F.cache\u002Fmatplotlib \\\n  -e TORCHINDUCTOR_CACHE_DIR=\u003Coutput_dir>\u002F.cache\u002Ftorchinductor \\\n  -e XDG_CACHE_HOME=\u003Coutput_dir>\u002F.cache\u002Fxdg \\\n  -v \u003Cdata_root>:\u003Cdata_root>:ro \\\n  -v \u003Coutput_dir>:\u003Coutput_dir> \\\n  \u003Ccontainer> \\\n  depth_net \u003Caction> -e \u003Cspec.yaml>\n```\n\nWithout `--user \"$(id -u):$(id -g)\"` the container writes outputs as `nobody:nogroup`, blocking host-side cleanup \u002F retry.\n\n### Step 5 — Verify\n\n- Container exit code 0\n- `status.json` `kpi` block populated\n- For `train`: inspect per-step `train_loss` directly (the entrypoint reports `Execution status: PASS` even when loss is NaN)\n- For `evaluate`: rely on `epe` \u002F `bp1` \u002F `bp2` \u002F `bp3` \u002F `d1` \u002F `rmse` (the evaluator also emits `abs_rel` \u002F `sq_rel` \u002F `rmse_log` which are non-meaningful for stereo — see `references\u002Fparameters-foundation-stereo.md`)\n- For `inference`: artifacts under `results_dir`\n\nFor TAO Deploy TensorRT actions (`gen_trt_engine`, TensorRT `evaluate`, and TensorRT `inference`), read `references\u002Ftao-deploy-foundation-stereo.md` first. Deploy spec templates live in this skill's `references\u002F` folder with the `spec_template_deploy_*.yaml` prefix.\n\n## Training Requirements\n\n- **Monitoring metric:** val\u002Floss\n- **Eval dataset:** optional. Val dataset configured via `dataset.val_dataset.data_sources` (each entry needs `data_file` and `dataset_name`).\n\nSee `references\u002Fspec-overrides-foundation-stereo.md` for the per-action dataset-requirements table and every action's mandatory data-source override block.\n\n## Parameters, Metrics, Multi-GPU, Export\u002FTRT, Hardware\n\nSee `references\u002Fparameters-foundation-stereo.md` for the full Important Parameters list (incl. `model.encoder` `vits` override, `model.max_disparity` default 416, `model.volume_dim` no-op note, `dataset.baseline`, `dataset.focal_x`, `train.precision`, `export.batch_size`), the Evaluation Metrics table, Multi-GPU \u002F Multi-Node launch keys, Export \u002F TRT Defaults (`opset_version`\u002F`on_cpu` pairing, NGC 576×960 settings), and Hardware requirements.\n\n## Error Patterns and Troubleshooting\n\nSee `references\u002Ftroubleshooting-foundation-stereo.md` for disparity overflow, smoke-test shape mismatch, missing pretrained paths, the `encoder` \u002F `dataset_name` struct errors, the `depth_net_stereo: not found` entrypoint note, the pyt-vs-deploy `crop_size` discussion, and the deploy `evaluate` scalar-conversion failure.\n\n## Spec Param \u002F Parent Model Inference\n\nSee `references\u002Fcheckpoint-inference-mappings-foundation-stereo.md` for the checkpoint-resolution rules (`model_epoch_\u003Cepoch>_step_\u003Cstep>.pth`, `dn_model_latest.pth` policy), the absence of parent PyT `gen_trt_engine`, and the full per-action inference-mapping table from `depth_net_stereo.config.json` (including `parent_model` \u002F `parent_job_id` resolution).\n\n## Deployment\n\n- [tao-deploy-foundation-stereo](references\u002Ftao-deploy-foundation-stereo.md)\n",{"data":35,"body":45},{"name":4,"description":6,"license":26,"compatibility":36,"metadata":37,"allowed-tools":40,"tags":41},"Requires docker + nvidia-container-toolkit.",{"version":38,"author":39},"0.1.0","NVIDIA Corporation","Read Bash",[42,43,44],"stereo","depth","estimation",{"type":46,"children":47},"root",[48,57,63,85,114,172,213,220,332,363,369,376,381,428,440,453,459,472,571,583,592,603,862,926,947,960,1167,1229,1300,1306,1319,1387,1457,1493,1540,1546,1567,1702,1711,1732,1738,1892,1937,1943,1986,1998,2004,2091,2097,2144,2150,2207,2213,2225],{"type":49,"tag":50,"props":51,"children":53},"element","h1",{"id":52},"depth-net-stereo",[54],{"type":55,"value":56},"text","Depth Net Stereo",{"type":49,"tag":58,"props":59,"children":60},"p",{},[61],{"type":55,"value":62},"Stereo depth estimation using FoundationStereo architecture. Predicts disparity maps from stereo image pairs for 3D reconstruction.",{"type":49,"tag":58,"props":64,"children":65},{},[66,68,75,77,83],{"type":55,"value":67},"Uses pretrained Depth Anything v2 and EdgeNeXt encoders. Set ",{"type":49,"tag":69,"props":70,"children":72},"code",{"className":71},[],[73],{"type":55,"value":74},"model.stereo_backbone.depth_anything_v2_pretrained_path",{"type":55,"value":76}," and ",{"type":49,"tag":69,"props":78,"children":80},{"className":79},[],[81],{"type":55,"value":82},"model.stereo_backbone.edgenext_pretrained_path",{"type":55,"value":84},".",{"type":49,"tag":58,"props":86,"children":87},{},[88,90,96,98,104,106,112],{"type":55,"value":89},"The mono and stereo skills both invoke the unified TAO ",{"type":49,"tag":69,"props":91,"children":93},{"className":92},[],[94],{"type":55,"value":95},"depth_net",{"type":55,"value":97}," CLI inside the container; the mono\u002Fstereo family is selected via ",{"type":49,"tag":69,"props":99,"children":101},{"className":100},[],[102],{"type":55,"value":103},"model.model_type",{"type":55,"value":105}," (e.g., ",{"type":49,"tag":69,"props":107,"children":109},{"className":108},[],[110],{"type":55,"value":111},"FoundationStereo",{"type":55,"value":113},").",{"type":49,"tag":58,"props":115,"children":116},{},[117,119,125,127,133,134,140,141,147,149,155,157,162,164,170],{"type":55,"value":118},"PyT actions packaged by this model skill: ",{"type":49,"tag":69,"props":120,"children":122},{"className":121},[],[123],{"type":55,"value":124},"train",{"type":55,"value":126},", ",{"type":49,"tag":69,"props":128,"children":130},{"className":129},[],[131],{"type":55,"value":132},"evaluate",{"type":55,"value":126},{"type":49,"tag":69,"props":135,"children":137},{"className":136},[],[138],{"type":55,"value":139},"inference",{"type":55,"value":126},{"type":49,"tag":69,"props":142,"children":144},{"className":143},[],[145],{"type":55,"value":146},"export",{"type":55,"value":148},", and ",{"type":49,"tag":69,"props":150,"children":152},{"className":151},[],[153],{"type":55,"value":154},"quantize",{"type":55,"value":156},". The PyT ",{"type":49,"tag":69,"props":158,"children":160},{"className":159},[],[161],{"type":55,"value":95},{"type":55,"value":163}," entrypoint does not accept a ",{"type":49,"tag":69,"props":165,"children":167},{"className":166},[],[168],{"type":55,"value":169},"gen_trt_engine",{"type":55,"value":171}," action in the current TAO image; build TensorRT engines only through the deploy workflow.",{"type":49,"tag":58,"props":173,"children":174},{},[175,177,182,184,189,191,196,198,204,206,212],{"type":55,"value":176},"For TAO Deploy TensorRT actions (",{"type":49,"tag":69,"props":178,"children":180},{"className":179},[],[181],{"type":55,"value":169},{"type":55,"value":183},", TensorRT ",{"type":49,"tag":69,"props":185,"children":187},{"className":186},[],[188],{"type":55,"value":132},{"type":55,"value":190},", and TensorRT ",{"type":49,"tag":69,"props":192,"children":194},{"className":193},[],[195],{"type":55,"value":139},{"type":55,"value":197},"), read ",{"type":49,"tag":69,"props":199,"children":201},{"className":200},[],[202],{"type":55,"value":203},"references\u002Ftao-deploy-foundation-stereo.md",{"type":55,"value":205}," first. The deploy spec template lives in this skill's ",{"type":49,"tag":69,"props":207,"children":209},{"className":208},[],[210],{"type":55,"value":211},"references\u002Fspec_template_deploy.yaml",{"type":55,"value":84},{"type":49,"tag":214,"props":215,"children":217},"h2",{"id":216},"train-action-policy",[218],{"type":55,"value":219},"Train Action Policy",{"type":49,"tag":58,"props":221,"children":222},{},[223,225,231,233,239,241,247,249,255,257,263,265,271,273,278,279,285,287,293,294,300,302,308,310,316,318,323,325,330],{"type":55,"value":224},"This model is AutoML-enabled at the model layer. Before handling any train-stage request, read ",{"type":49,"tag":69,"props":226,"children":228},{"className":227},[],[229],{"type":55,"value":230},"references\u002Fskill_info.yaml",{"type":55,"value":232}," and resolve the run override from either an explicit ",{"type":49,"tag":69,"props":234,"children":236},{"className":235},[],[237],{"type":55,"value":238},"automl_policy",{"type":55,"value":240}," value or the user's workflow request. Use ",{"type":49,"tag":69,"props":242,"children":244},{"className":243},[],[245],{"type":55,"value":246},"automl_policy: on",{"type":55,"value":248}," by default and only expose ",{"type":49,"tag":69,"props":250,"children":252},{"className":251},[],[253],{"type":55,"value":254},"on",{"type":55,"value":256}," \u002F ",{"type":49,"tag":69,"props":258,"children":260},{"className":259},[],[261],{"type":55,"value":262},"off",{"type":55,"value":264}," in new launch prompts. Treat phrases like \"turn off AutoML\", \"disable AutoML\", \"no HPO\", or \"plain training\" as ",{"type":49,"tag":69,"props":266,"children":268},{"className":267},[],[269],{"type":55,"value":270},"automl_policy: off",{"type":55,"value":272}," for this run only. When ",{"type":49,"tag":69,"props":274,"children":276},{"className":275},[],[277],{"type":55,"value":246},{"type":55,"value":126},{"type":49,"tag":69,"props":280,"children":282},{"className":281},[],[283],{"type":55,"value":284},"automl_enabled: true",{"type":55,"value":286},", and both ",{"type":49,"tag":69,"props":288,"children":290},{"className":289},[],[291],{"type":55,"value":292},"schemas\u002Ftrain.schema.json",{"type":55,"value":76},{"type":49,"tag":69,"props":295,"children":297},{"className":296},[],[298],{"type":55,"value":299},"references\u002Fspec_template_train.yaml",{"type":55,"value":301}," are packaged, route the train action through ",{"type":49,"tag":69,"props":303,"children":305},{"className":304},[],[306],{"type":55,"value":307},"tao-skill-bank:tao-run-automl",{"type":55,"value":309}," by default with this model's ",{"type":49,"tag":69,"props":311,"children":313},{"className":312},[],[314],{"type":55,"value":315},"skill_dir",{"type":55,"value":317},". Preserve workflow\u002Fapplication overrides for datasets, specs, output directories, GPU\u002Fplatform settings, parent checkpoints, and ",{"type":49,"tag":69,"props":319,"children":321},{"className":320},[],[322],{"type":55,"value":238},{"type":55,"value":324},". Use direct model training only when ",{"type":49,"tag":69,"props":326,"children":328},{"className":327},[],[329],{"type":55,"value":270},{"type":55,"value":331}," or the packaged train schema\u002Ftemplate is missing; in the missing-schema case, report that AutoML is enabled but not runnable for this model until schemas are generated.",{"type":49,"tag":58,"props":333,"children":334},{},[335,337,342,343,348,349,354,356,361],{"type":55,"value":336},"Non-train actions such as ",{"type":49,"tag":69,"props":338,"children":340},{"className":339},[],[341],{"type":55,"value":132},{"type":55,"value":126},{"type":49,"tag":69,"props":344,"children":346},{"className":345},[],[347],{"type":55,"value":139},{"type":55,"value":126},{"type":49,"tag":69,"props":350,"children":352},{"className":351},[],[353],{"type":55,"value":146},{"type":55,"value":355},", and deploy flows stay in this model skill. The per-run ",{"type":49,"tag":69,"props":357,"children":359},{"className":358},[],[360],{"type":55,"value":238},{"type":55,"value":362}," override does not change model metadata.",{"type":49,"tag":214,"props":364,"children":366},{"id":365},"workflow",[367],{"type":55,"value":368},"Workflow",{"type":49,"tag":370,"props":371,"children":373},"h3",{"id":372},"prerequisites-data-accessibility",[374],{"type":55,"value":375},"Prerequisites — data accessibility",{"type":49,"tag":58,"props":377,"children":378},{},[379],{"type":55,"value":380},"Your dataset (left + right images + GT disparity) must be reachable from inside the container:",{"type":49,"tag":382,"props":383,"children":384},"ul",{},[385,412],{"type":49,"tag":386,"props":387,"children":388},"li",{},[389,395,397,403,404,410],{"type":49,"tag":390,"props":391,"children":392},"strong",{},[393],{"type":55,"value":394},"SDK runner",{"type":55,"value":396},": place files at the S3 paths the runner resolves (the ",{"type":49,"tag":69,"props":398,"children":400},{"className":399},[],[401],{"type":55,"value":402},"S3_TRAIN",{"type":55,"value":256},{"type":49,"tag":69,"props":405,"children":407},{"className":406},[],[408],{"type":55,"value":409},"S3_EVAL",{"type":55,"value":411}," placeholders shown in the spec overrides). The runner handles S3 → container-path mounting transparently.",{"type":49,"tag":386,"props":413,"children":414},{},[415,426],{"type":49,"tag":390,"props":416,"children":417},{},[418,420],{"type":55,"value":419},"Direct ",{"type":49,"tag":69,"props":421,"children":423},{"className":422},[],[424],{"type":55,"value":425},"docker run",{"type":55,"value":427}," (e.g. local testing): mount the host dataset root read-only at the same in-container path:",{"type":49,"tag":429,"props":430,"children":434},"pre",{"className":431,"code":433,"language":55},[432],"language-text","docker run ... -v \u003Chost_data_root>:\u003Chost_data_root>:ro \u003Ccontainer> ...\n",[435],{"type":49,"tag":69,"props":436,"children":438},{"__ignoreMap":437},"",[439],{"type":55,"value":433},{"type":49,"tag":58,"props":441,"children":442},{},[443,445,451],{"type":55,"value":444},"The same accessibility requirement applies to the ",{"type":49,"tag":69,"props":446,"children":448},{"className":447},[],[449],{"type":55,"value":450},"\u003Coutput_dir>",{"type":55,"value":452}," written by all actions.",{"type":49,"tag":370,"props":454,"children":456},{"id":455},"step-1-annotation-file",[457],{"type":55,"value":458},"Step 1 — Annotation file",{"type":49,"tag":58,"props":460,"children":461},{},[462,464,470],{"type":55,"value":463},"Per-line annotation file referenced by ",{"type":49,"tag":69,"props":465,"children":467},{"className":466},[],[468],{"type":55,"value":469},"data_sources[*].data_file",{"type":55,"value":471},":",{"type":49,"tag":473,"props":474,"children":475},"table",{},[476,500],{"type":49,"tag":477,"props":478,"children":479},"thead",{},[480],{"type":49,"tag":481,"props":482,"children":483},"tr",{},[484,490,495],{"type":49,"tag":485,"props":486,"children":487},"th",{},[488],{"type":55,"value":489},"Columns",{"type":49,"tag":485,"props":491,"children":492},{},[493],{"type":55,"value":494},"Format",{"type":49,"tag":485,"props":496,"children":497},{},[498],{"type":55,"value":499},"Use",{"type":49,"tag":501,"props":502,"children":503},"tbody",{},[504,527,549],{"type":49,"tag":481,"props":505,"children":506},{},[507,513,522],{"type":49,"tag":508,"props":509,"children":510},"td",{},[511],{"type":55,"value":512},"2",{"type":49,"tag":508,"props":514,"children":515},{},[516],{"type":49,"tag":69,"props":517,"children":519},{"className":518},[],[520],{"type":55,"value":521},"\u003Cleft> \u003Cright>",{"type":49,"tag":508,"props":523,"children":524},{},[525],{"type":55,"value":526},"Stereo inference (no GT)",{"type":49,"tag":481,"props":528,"children":529},{},[530,535,544],{"type":49,"tag":508,"props":531,"children":532},{},[533],{"type":55,"value":534},"3",{"type":49,"tag":508,"props":536,"children":537},{},[538],{"type":49,"tag":69,"props":539,"children":541},{"className":540},[],[542],{"type":55,"value":543},"\u003Cleft> \u003Cright> \u003Cdisparity>",{"type":49,"tag":508,"props":545,"children":546},{},[547],{"type":55,"value":548},"Stereo with GT",{"type":49,"tag":481,"props":550,"children":551},{},[552,557,566],{"type":49,"tag":508,"props":553,"children":554},{},[555],{"type":55,"value":556},"4",{"type":49,"tag":508,"props":558,"children":559},{},[560],{"type":49,"tag":69,"props":561,"children":563},{"className":562},[],[564],{"type":55,"value":565},"\u003Cleft> \u003Cright> \u003Cdisparity> \u003Cocclusion_mask>",{"type":49,"tag":508,"props":567,"children":568},{},[569],{"type":55,"value":570},"Stereo with GT and occlusion mask",{"type":49,"tag":58,"props":572,"children":573},{},[574,576,582],{"type":55,"value":575},"If you already have one, point to it. Otherwise generate via ",{"type":49,"tag":69,"props":577,"children":579},{"className":578},[],[580],{"type":55,"value":581},"depth_net convert",{"type":55,"value":471},{"type":49,"tag":429,"props":584,"children":587},{"className":585,"code":586,"language":55},[432],"depth_net convert -e \u003Cconvert_spec.yaml>\n",[588],{"type":49,"tag":69,"props":589,"children":590},{"__ignoreMap":437},[591],{"type":55,"value":586},{"type":49,"tag":58,"props":593,"children":594},{},[595,601],{"type":49,"tag":69,"props":596,"children":598},{"className":597},[],[599],{"type":55,"value":600},"convert_spec.yaml",{"type":55,"value":602}," template (stereo):",{"type":49,"tag":429,"props":604,"children":608},{"className":605,"code":606,"language":607,"meta":437,"style":437},"language-yaml shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","results_dir: \u003Cdirectory where generated annotation files are written>\ndata_root: \u003Cdirectory whose immediate children are scene folders that contain your image+depth files; convert walks data_root recursively but expects per-scene subdirectories at one level below>\nimage_dir_pattern: [\u003Csubstring matching left image paths>]\nright_dir_pattern: [\u003Csubstring matching right image paths>]\ndepth_dir_pattern: [\u003Csubstring matching GT disparity paths>]\nnocc_dir_pattern: []                 # optional, occlusion mask paths\nimage_extension: '.png'  # always include the leading dot\ndepth_extension: '.png'  # form must match image_extension (the swap is a substring replace)\nnocc_extension: ''\nsplit_ratio: 0.0        # 0.0\u002F1.0 = test-only; 0.8 = 80\u002F20 train+val\n","yaml",[609],{"type":49,"tag":69,"props":610,"children":611},{"__ignoreMap":437},[612,635,653,681,707,733,757,790,820,838],{"type":49,"tag":613,"props":614,"children":617},"span",{"class":615,"line":616},"line",1,[618,624,629],{"type":49,"tag":613,"props":619,"children":621},{"style":620},"--shiki-light:#E53935;--shiki-default:#F07178;--shiki-dark:#F07178",[622],{"type":55,"value":623},"results_dir",{"type":49,"tag":613,"props":625,"children":627},{"style":626},"--shiki-light:#39ADB5;--shiki-default:#89DDFF;--shiki-dark:#89DDFF",[628],{"type":55,"value":471},{"type":49,"tag":613,"props":630,"children":632},{"style":631},"--shiki-light:#91B859;--shiki-default:#C3E88D;--shiki-dark:#C3E88D",[633],{"type":55,"value":634}," \u003Cdirectory where generated annotation files are written>\n",{"type":49,"tag":613,"props":636,"children":638},{"class":615,"line":637},2,[639,644,648],{"type":49,"tag":613,"props":640,"children":641},{"style":620},[642],{"type":55,"value":643},"data_root",{"type":49,"tag":613,"props":645,"children":646},{"style":626},[647],{"type":55,"value":471},{"type":49,"tag":613,"props":649,"children":650},{"style":631},[651],{"type":55,"value":652}," \u003Cdirectory whose immediate children are scene folders that contain your image+depth files; convert walks data_root recursively but expects per-scene subdirectories at one level below>\n",{"type":49,"tag":613,"props":654,"children":656},{"class":615,"line":655},3,[657,662,666,671,676],{"type":49,"tag":613,"props":658,"children":659},{"style":620},[660],{"type":55,"value":661},"image_dir_pattern",{"type":49,"tag":613,"props":663,"children":664},{"style":626},[665],{"type":55,"value":471},{"type":49,"tag":613,"props":667,"children":668},{"style":626},[669],{"type":55,"value":670}," [",{"type":49,"tag":613,"props":672,"children":673},{"style":631},[674],{"type":55,"value":675},"\u003Csubstring matching left image paths>",{"type":49,"tag":613,"props":677,"children":678},{"style":626},[679],{"type":55,"value":680},"]\n",{"type":49,"tag":613,"props":682,"children":684},{"class":615,"line":683},4,[685,690,694,698,703],{"type":49,"tag":613,"props":686,"children":687},{"style":620},[688],{"type":55,"value":689},"right_dir_pattern",{"type":49,"tag":613,"props":691,"children":692},{"style":626},[693],{"type":55,"value":471},{"type":49,"tag":613,"props":695,"children":696},{"style":626},[697],{"type":55,"value":670},{"type":49,"tag":613,"props":699,"children":700},{"style":631},[701],{"type":55,"value":702},"\u003Csubstring matching right image paths>",{"type":49,"tag":613,"props":704,"children":705},{"style":626},[706],{"type":55,"value":680},{"type":49,"tag":613,"props":708,"children":710},{"class":615,"line":709},5,[711,716,720,724,729],{"type":49,"tag":613,"props":712,"children":713},{"style":620},[714],{"type":55,"value":715},"depth_dir_pattern",{"type":49,"tag":613,"props":717,"children":718},{"style":626},[719],{"type":55,"value":471},{"type":49,"tag":613,"props":721,"children":722},{"style":626},[723],{"type":55,"value":670},{"type":49,"tag":613,"props":725,"children":726},{"style":631},[727],{"type":55,"value":728},"\u003Csubstring matching GT disparity paths>",{"type":49,"tag":613,"props":730,"children":731},{"style":626},[732],{"type":55,"value":680},{"type":49,"tag":613,"props":734,"children":736},{"class":615,"line":735},6,[737,742,746,751],{"type":49,"tag":613,"props":738,"children":739},{"style":620},[740],{"type":55,"value":741},"nocc_dir_pattern",{"type":49,"tag":613,"props":743,"children":744},{"style":626},[745],{"type":55,"value":471},{"type":49,"tag":613,"props":747,"children":748},{"style":626},[749],{"type":55,"value":750}," []",{"type":49,"tag":613,"props":752,"children":754},{"style":753},"--shiki-light:#90A4AE;--shiki-light-font-style:italic;--shiki-default:#546E7A;--shiki-default-font-style:italic;--shiki-dark:#676E95;--shiki-dark-font-style:italic",[755],{"type":55,"value":756},"                 # optional, occlusion mask paths\n",{"type":49,"tag":613,"props":758,"children":760},{"class":615,"line":759},7,[761,766,770,775,780,785],{"type":49,"tag":613,"props":762,"children":763},{"style":620},[764],{"type":55,"value":765},"image_extension",{"type":49,"tag":613,"props":767,"children":768},{"style":626},[769],{"type":55,"value":471},{"type":49,"tag":613,"props":771,"children":772},{"style":626},[773],{"type":55,"value":774}," '",{"type":49,"tag":613,"props":776,"children":777},{"style":631},[778],{"type":55,"value":779},".png",{"type":49,"tag":613,"props":781,"children":782},{"style":626},[783],{"type":55,"value":784},"'",{"type":49,"tag":613,"props":786,"children":787},{"style":753},[788],{"type":55,"value":789},"  # always include the leading dot\n",{"type":49,"tag":613,"props":791,"children":793},{"class":615,"line":792},8,[794,799,803,807,811,815],{"type":49,"tag":613,"props":795,"children":796},{"style":620},[797],{"type":55,"value":798},"depth_extension",{"type":49,"tag":613,"props":800,"children":801},{"style":626},[802],{"type":55,"value":471},{"type":49,"tag":613,"props":804,"children":805},{"style":626},[806],{"type":55,"value":774},{"type":49,"tag":613,"props":808,"children":809},{"style":631},[810],{"type":55,"value":779},{"type":49,"tag":613,"props":812,"children":813},{"style":626},[814],{"type":55,"value":784},{"type":49,"tag":613,"props":816,"children":817},{"style":753},[818],{"type":55,"value":819},"  # form must match image_extension (the swap is a substring replace)\n",{"type":49,"tag":613,"props":821,"children":823},{"class":615,"line":822},9,[824,829,833],{"type":49,"tag":613,"props":825,"children":826},{"style":620},[827],{"type":55,"value":828},"nocc_extension",{"type":49,"tag":613,"props":830,"children":831},{"style":626},[832],{"type":55,"value":471},{"type":49,"tag":613,"props":834,"children":835},{"style":626},[836],{"type":55,"value":837}," ''\n",{"type":49,"tag":613,"props":839,"children":841},{"class":615,"line":840},10,[842,847,851,857],{"type":49,"tag":613,"props":843,"children":844},{"style":620},[845],{"type":55,"value":846},"split_ratio",{"type":49,"tag":613,"props":848,"children":849},{"style":626},[850],{"type":55,"value":471},{"type":49,"tag":613,"props":852,"children":854},{"style":853},"--shiki-light:#F76D47;--shiki-default:#F78C6C;--shiki-dark:#F78C6C",[855],{"type":55,"value":856}," 0.0",{"type":49,"tag":613,"props":858,"children":859},{"style":753},[860],{"type":55,"value":861},"        # 0.0\u002F1.0 = test-only; 0.8 = 80\u002F20 train+val\n",{"type":49,"tag":58,"props":863,"children":864},{},[865,871,873,878,880,886,888,893,895,901,903,909,911,917,918,924],{"type":49,"tag":69,"props":866,"children":868},{"className":867},[],[869],{"type":55,"value":870},"convert",{"type":55,"value":872}," walks ",{"type":49,"tag":69,"props":874,"children":876},{"className":875},[],[877],{"type":55,"value":643},{"type":55,"value":879}," recursively, selects paths whose path-string contains ",{"type":49,"tag":881,"props":882,"children":883},"em",{},[884],{"type":55,"value":885},"all",{"type":55,"value":887}," substrings in ",{"type":49,"tag":69,"props":889,"children":891},{"className":890},[],[892],{"type":55,"value":661},{"type":55,"value":894}," (AND-filter), then derives right \u002F depth \u002F mask paths by replacing ",{"type":49,"tag":69,"props":896,"children":898},{"className":897},[],[899],{"type":55,"value":900},"image_dir_pattern[0]",{"type":55,"value":902}," with the corresponding pattern's first element plus extension swap. Inspect your dataset's directory layout and identify the substrings distinguishing left, right, and GT (e.g. ",{"type":49,"tag":69,"props":904,"children":906},{"className":905},[],[907],{"type":55,"value":908},"im0",{"type":55,"value":910}," vs ",{"type":49,"tag":69,"props":912,"children":914},{"className":913},[],[915],{"type":55,"value":916},"im1",{"type":55,"value":910},{"type":49,"tag":69,"props":919,"children":921},{"className":920},[],[922],{"type":55,"value":923},"disp0GT",{"type":55,"value":925}," for Middlebury).",{"type":49,"tag":370,"props":927,"children":929},{"id":928},"step-2-pair-model_type-and-dataset_name-based-on-your-data",[930,932,938,939,945],{"type":55,"value":931},"Step 2 — Pair ",{"type":49,"tag":69,"props":933,"children":935},{"className":934},[],[936],{"type":55,"value":937},"model_type",{"type":55,"value":76},{"type":49,"tag":69,"props":940,"children":942},{"className":941},[],[943],{"type":55,"value":944},"dataset_name",{"type":55,"value":946}," based on your data",{"type":49,"tag":58,"props":948,"children":949},{},[950,952,958],{"type":55,"value":951},"Prefer the dataset-specific class when your layout matches a supported one — it applies class-specific path conventions, evaluation crops, and (where applicable) occlusion-mask handling. Fall back to ",{"type":49,"tag":69,"props":953,"children":955},{"className":954},[],[956],{"type":55,"value":957},"GenericDataset",{"type":55,"value":959}," only for layouts that do not match any registered class.",{"type":49,"tag":473,"props":961,"children":962},{},[963,990],{"type":49,"tag":477,"props":964,"children":965},{},[966],{"type":49,"tag":481,"props":967,"children":968},{},[969,974,982],{"type":49,"tag":485,"props":970,"children":971},{},[972],{"type":55,"value":973},"Data category",{"type":49,"tag":485,"props":975,"children":976},{},[977],{"type":49,"tag":69,"props":978,"children":980},{"className":979},[],[981],{"type":55,"value":937},{"type":49,"tag":485,"props":983,"children":984},{},[985],{"type":49,"tag":69,"props":986,"children":988},{"className":987},[],[989],{"type":55,"value":944},{"type":49,"tag":501,"props":991,"children":992},{},[993,1018,1043,1068,1093,1118,1143],{"type":49,"tag":481,"props":994,"children":995},{},[996,1001,1009],{"type":49,"tag":508,"props":997,"children":998},{},[999],{"type":55,"value":1000},"Middlebury data",{"type":49,"tag":508,"props":1002,"children":1003},{},[1004],{"type":49,"tag":69,"props":1005,"children":1007},{"className":1006},[],[1008],{"type":55,"value":111},{"type":49,"tag":508,"props":1010,"children":1011},{},[1012],{"type":49,"tag":69,"props":1013,"children":1015},{"className":1014},[],[1016],{"type":55,"value":1017},"Middlebury",{"type":49,"tag":481,"props":1019,"children":1020},{},[1021,1026,1034],{"type":49,"tag":508,"props":1022,"children":1023},{},[1024],{"type":55,"value":1025},"KITTI data",{"type":49,"tag":508,"props":1027,"children":1028},{},[1029],{"type":49,"tag":69,"props":1030,"children":1032},{"className":1031},[],[1033],{"type":55,"value":111},{"type":49,"tag":508,"props":1035,"children":1036},{},[1037],{"type":49,"tag":69,"props":1038,"children":1040},{"className":1039},[],[1041],{"type":55,"value":1042},"Kitti",{"type":49,"tag":481,"props":1044,"children":1045},{},[1046,1051,1059],{"type":49,"tag":508,"props":1047,"children":1048},{},[1049],{"type":55,"value":1050},"ETH3D data",{"type":49,"tag":508,"props":1052,"children":1053},{},[1054],{"type":49,"tag":69,"props":1055,"children":1057},{"className":1056},[],[1058],{"type":55,"value":111},{"type":49,"tag":508,"props":1060,"children":1061},{},[1062],{"type":49,"tag":69,"props":1063,"children":1065},{"className":1064},[],[1066],{"type":55,"value":1067},"Eth3d",{"type":49,"tag":481,"props":1069,"children":1070},{},[1071,1076,1084],{"type":49,"tag":508,"props":1072,"children":1073},{},[1074],{"type":55,"value":1075},"FSD synthetic data",{"type":49,"tag":508,"props":1077,"children":1078},{},[1079],{"type":49,"tag":69,"props":1080,"children":1082},{"className":1081},[],[1083],{"type":55,"value":111},{"type":49,"tag":508,"props":1085,"children":1086},{},[1087],{"type":49,"tag":69,"props":1088,"children":1090},{"className":1089},[],[1091],{"type":55,"value":1092},"FSD",{"type":49,"tag":481,"props":1094,"children":1095},{},[1096,1101,1109],{"type":49,"tag":508,"props":1097,"children":1098},{},[1099],{"type":55,"value":1100},"IsaacReal synthetic data",{"type":49,"tag":508,"props":1102,"children":1103},{},[1104],{"type":49,"tag":69,"props":1105,"children":1107},{"className":1106},[],[1108],{"type":55,"value":111},{"type":49,"tag":508,"props":1110,"children":1111},{},[1112],{"type":49,"tag":69,"props":1113,"children":1115},{"className":1114},[],[1116],{"type":55,"value":1117},"IsaacRealDataset",{"type":49,"tag":481,"props":1119,"children":1120},{},[1121,1126,1134],{"type":49,"tag":508,"props":1122,"children":1123},{},[1124],{"type":55,"value":1125},"Crestereo synthetic data",{"type":49,"tag":508,"props":1127,"children":1128},{},[1129],{"type":49,"tag":69,"props":1130,"children":1132},{"className":1131},[],[1133],{"type":55,"value":111},{"type":49,"tag":508,"props":1135,"children":1136},{},[1137],{"type":49,"tag":69,"props":1138,"children":1140},{"className":1139},[],[1141],{"type":55,"value":1142},"Crestereo",{"type":49,"tag":481,"props":1144,"children":1145},{},[1146,1151,1159],{"type":49,"tag":508,"props":1147,"children":1148},{},[1149],{"type":55,"value":1150},"Other \u002F non-canonical layout",{"type":49,"tag":508,"props":1152,"children":1153},{},[1154],{"type":49,"tag":69,"props":1155,"children":1157},{"className":1156},[],[1158],{"type":55,"value":111},{"type":49,"tag":508,"props":1160,"children":1161},{},[1162],{"type":49,"tag":69,"props":1163,"children":1165},{"className":1164},[],[1166],{"type":55,"value":957},{"type":49,"tag":58,"props":1168,"children":1169},{},[1170,1172,1177,1179,1185,1187,1192,1193,1198,1199,1204,1205,1210,1211,1216,1217,1222,1223,1228],{"type":55,"value":1171},"Valid ",{"type":49,"tag":69,"props":1173,"children":1175},{"className":1174},[],[1176],{"type":55,"value":944},{"type":55,"value":1178}," values for stereo ",{"type":49,"tag":69,"props":1180,"children":1182},{"className":1181},[],[1183],{"type":55,"value":1184},"data_sources",{"type":55,"value":1186}," (case-insensitive): ",{"type":49,"tag":69,"props":1188,"children":1190},{"className":1189},[],[1191],{"type":55,"value":1092},{"type":55,"value":126},{"type":49,"tag":69,"props":1194,"children":1196},{"className":1195},[],[1197],{"type":55,"value":1117},{"type":55,"value":126},{"type":49,"tag":69,"props":1200,"children":1202},{"className":1201},[],[1203],{"type":55,"value":1142},{"type":55,"value":126},{"type":49,"tag":69,"props":1206,"children":1208},{"className":1207},[],[1209],{"type":55,"value":1017},{"type":55,"value":126},{"type":49,"tag":69,"props":1212,"children":1214},{"className":1213},[],[1215],{"type":55,"value":1067},{"type":55,"value":126},{"type":49,"tag":69,"props":1218,"children":1220},{"className":1219},[],[1221],{"type":55,"value":1042},{"type":55,"value":126},{"type":49,"tag":69,"props":1224,"children":1226},{"className":1225},[],[1227],{"type":55,"value":957},{"type":55,"value":84},{"type":49,"tag":58,"props":1230,"children":1231},{},[1232,1234,1239,1241,1246,1248,1253,1255,1261,1263,1268,1269,1274,1275,1280,1281,1286,1287,1292,1293,1298],{"type":55,"value":1233},"The same ",{"type":49,"tag":69,"props":1235,"children":1237},{"className":1236},[],[1238],{"type":55,"value":944},{"type":55,"value":1240}," value applies across train and evaluate actions (all of which use 3-column or 4-column annotations with GT disparity). The deploy-side ",{"type":49,"tag":69,"props":1242,"children":1244},{"className":1243},[],[1245],{"type":55,"value":132},{"type":55,"value":1247}," action follows the same rule — see ",{"type":49,"tag":69,"props":1249,"children":1251},{"className":1250},[],[1252],{"type":55,"value":203},{"type":55,"value":1254},". For inference with 2-column annotations (left + right, no GT), use ",{"type":49,"tag":69,"props":1256,"children":1258},{"className":1257},[],[1259],{"type":55,"value":1260},"dataset_name: GenericDataset",{"type":55,"value":1262}," regardless of data layout — the dataset-specific classes (",{"type":49,"tag":69,"props":1264,"children":1266},{"className":1265},[],[1267],{"type":55,"value":1017},{"type":55,"value":256},{"type":49,"tag":69,"props":1270,"children":1272},{"className":1271},[],[1273],{"type":55,"value":1042},{"type":55,"value":256},{"type":49,"tag":69,"props":1276,"children":1278},{"className":1277},[],[1279],{"type":55,"value":1067},{"type":55,"value":256},{"type":49,"tag":69,"props":1282,"children":1284},{"className":1283},[],[1285],{"type":55,"value":1092},{"type":55,"value":256},{"type":49,"tag":69,"props":1288,"children":1290},{"className":1289},[],[1291],{"type":55,"value":1117},{"type":55,"value":256},{"type":49,"tag":69,"props":1294,"children":1296},{"className":1295},[],[1297],{"type":55,"value":1142},{"type":55,"value":1299},") require 3-column input and reject 2-column annotations at the dataloader level. For inference with 3-column annotations (left + right + GT), the dataset-specific class is fine.",{"type":49,"tag":370,"props":1301,"children":1303},{"id":1302},"step-3-write-spec-yaml-from-the-spec-overrides",[1304],{"type":55,"value":1305},"Step 3 — Write spec yaml from the spec overrides",{"type":49,"tag":58,"props":1307,"children":1308},{},[1309,1311,1317],{"type":55,"value":1310},"Copy the action block from ",{"type":49,"tag":69,"props":1312,"children":1314},{"className":1313},[],[1315],{"type":55,"value":1316},"references\u002Fspec-overrides-foundation-stereo.md",{"type":55,"value":1318},". Replace:",{"type":49,"tag":382,"props":1320,"children":1321},{},[1322,1339,1350,1361],{"type":49,"tag":386,"props":1323,"children":1324},{},[1325,1330,1332,1337],{"type":49,"tag":69,"props":1326,"children":1328},{"className":1327},[],[1329],{"type":55,"value":103},{"type":55,"value":1331}," from Step 2 (typically ",{"type":49,"tag":69,"props":1333,"children":1335},{"className":1334},[],[1336],{"type":55,"value":111},{"type":55,"value":1338},")",{"type":49,"tag":386,"props":1340,"children":1341},{},[1342,1348],{"type":49,"tag":69,"props":1343,"children":1345},{"className":1344},[],[1346],{"type":55,"value":1347},"dataset.\u003C...>.data_sources[*].dataset_name",{"type":55,"value":1349}," from Step 2",{"type":49,"tag":386,"props":1351,"children":1352},{},[1353,1359],{"type":49,"tag":69,"props":1354,"children":1356},{"className":1355},[],[1357],{"type":55,"value":1358},"dataset.\u003C...>.data_sources[*].data_file",{"type":55,"value":1360}," with the path from Step 1",{"type":49,"tag":386,"props":1362,"children":1363},{},[1364,1366,1371,1373,1379,1381,1386],{"type":55,"value":1365},"For deploy-side ",{"type":49,"tag":69,"props":1367,"children":1369},{"className":1368},[],[1370],{"type":55,"value":132},{"type":55,"value":1372},": enforce ",{"type":49,"tag":69,"props":1374,"children":1376},{"className":1375},[],[1377],{"type":55,"value":1378},"dataset.test_dataset.batch_size: 1",{"type":55,"value":1380}," (see ",{"type":49,"tag":69,"props":1382,"children":1384},{"className":1383},[],[1385],{"type":55,"value":203},{"type":55,"value":113},{"type":49,"tag":58,"props":1388,"children":1389},{},[1390,1392,1398,1400,1406,1408,1414,1415,1421,1423,1428,1430,1435,1437,1442,1444,1450,1451,1456],{"type":55,"value":1391},"Shape consistency: the ",{"type":49,"tag":69,"props":1393,"children":1395},{"className":1394},[],[1396],{"type":55,"value":1397},"crop_size",{"type":55,"value":1399}," in ",{"type":49,"tag":69,"props":1401,"children":1403},{"className":1402},[],[1404],{"type":55,"value":1405},"dataset.test_dataset.augmentation.crop_size",{"type":55,"value":1407}," should match ",{"type":49,"tag":69,"props":1409,"children":1411},{"className":1410},[],[1412],{"type":55,"value":1413},"export.input_height",{"type":55,"value":256},{"type":49,"tag":69,"props":1416,"children":1418},{"className":1417},[],[1419],{"type":55,"value":1420},"input_width",{"type":55,"value":1422}," so the trained-model evaluator and the deploy-side TensorRT evaluator operate at the same shape. Note that ",{"type":49,"tag":69,"props":1424,"children":1426},{"className":1425},[],[1427],{"type":55,"value":1397},{"type":55,"value":1429}," is decorative on the pyt ",{"type":49,"tag":69,"props":1431,"children":1433},{"className":1432},[],[1434],{"type":55,"value":132},{"type":55,"value":1436}," path but authoritative on the deploy ",{"type":49,"tag":69,"props":1438,"children":1440},{"className":1439},[],[1441],{"type":55,"value":132},{"type":55,"value":1443}," side — see ",{"type":49,"tag":69,"props":1445,"children":1447},{"className":1446},[],[1448],{"type":55,"value":1449},"references\u002Ftroubleshooting-foundation-stereo.md",{"type":55,"value":76},{"type":49,"tag":69,"props":1452,"children":1454},{"className":1453},[],[1455],{"type":55,"value":203},{"type":55,"value":84},{"type":49,"tag":58,"props":1458,"children":1459},{},[1460,1462,1468,1470,1476,1477,1483,1485,1491],{"type":55,"value":1461},"Fresh-install smoke runs are validated at ",{"type":49,"tag":69,"props":1463,"children":1465},{"className":1464},[],[1466],{"type":55,"value":1467},"crop_size: [128, 128]",{"type":55,"value":1469}," with ",{"type":49,"tag":69,"props":1471,"children":1473},{"className":1472},[],[1474],{"type":55,"value":1475},"dataset.max_disparity: 128",{"type":55,"value":76},{"type":49,"tag":69,"props":1478,"children":1480},{"className":1479},[],[1481],{"type":55,"value":1482},"model.max_disparity: 128",{"type":55,"value":1484},". Avoid 112×112 crops and avoid setting ",{"type":49,"tag":69,"props":1486,"children":1488},{"className":1487},[],[1489],{"type":55,"value":1490},"max_disparity",{"type":55,"value":1492}," smaller than the square crop side for smoke tests: those combinations can fail inside FoundationStereo with feature-map or loss-mask shape mismatches before a checkpoint is produced.",{"type":49,"tag":58,"props":1494,"children":1495},{},[1496,1498,1503,1505,1510,1512,1518,1519,1524,1526,1531,1533,1538],{"type":55,"value":1497},"Data source overrides are ",{"type":49,"tag":390,"props":1499,"children":1500},{},[1501],{"type":55,"value":1502},"mandatory for every action",{"type":55,"value":1504},". Each ",{"type":49,"tag":69,"props":1506,"children":1508},{"className":1507},[],[1509],{"type":55,"value":1184},{"type":55,"value":1511}," entry is a dict with two mandatory fields: ",{"type":49,"tag":69,"props":1513,"children":1515},{"className":1514},[],[1516],{"type":55,"value":1517},"data_file",{"type":55,"value":76},{"type":49,"tag":69,"props":1520,"children":1522},{"className":1521},[],[1523],{"type":55,"value":944},{"type":55,"value":1525},". See ",{"type":49,"tag":69,"props":1527,"children":1529},{"className":1528},[],[1530],{"type":55,"value":1316},{"type":55,"value":1532}," for the per-action dataset-requirements table, every action's override block, and the ",{"type":49,"tag":69,"props":1534,"children":1536},{"className":1535},[],[1537],{"type":55,"value":154},{"type":55,"value":1539}," known-issue note.",{"type":49,"tag":370,"props":1541,"children":1543},{"id":1542},"step-4-run",[1544],{"type":55,"value":1545},"Step 4 — Run",{"type":49,"tag":58,"props":1547,"children":1548},{},[1549,1551,1557,1559,1565],{"type":55,"value":1550},"Create writable home\u002Fcache directories inside the mounted output path before using\n",{"type":49,"tag":69,"props":1552,"children":1554},{"className":1553},[],[1555],{"type":55,"value":1556},"--user",{"type":55,"value":1558},". Some TAO containers do not have an ",{"type":49,"tag":69,"props":1560,"children":1562},{"className":1561},[],[1563],{"type":55,"value":1564},"\u002Fetc\u002Fpasswd",{"type":55,"value":1566}," entry for the host UID,\nand PyTorch \u002F matplotlib need writable cache paths when running as that UID.",{"type":49,"tag":429,"props":1568,"children":1572},{"className":1569,"code":1570,"language":1571,"meta":437,"style":437},"language-bash shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","mkdir -p \u003Coutput_dir>\u002Fhome \\\n         \u003Coutput_dir>\u002F.cache\u002Fmatplotlib \\\n         \u003Coutput_dir>\u002F.cache\u002Ftorchinductor \\\n         \u003Coutput_dir>\u002F.cache\u002Fxdg\n","bash",[1573],{"type":49,"tag":69,"props":1574,"children":1575},{"__ignoreMap":437},[1576,1621,1650,1678],{"type":49,"tag":613,"props":1577,"children":1578},{"class":615,"line":616},[1579,1585,1590,1595,1600,1606,1611,1616],{"type":49,"tag":613,"props":1580,"children":1582},{"style":1581},"--shiki-light:#E2931D;--shiki-default:#FFCB6B;--shiki-dark:#FFCB6B",[1583],{"type":55,"value":1584},"mkdir",{"type":49,"tag":613,"props":1586,"children":1587},{"style":631},[1588],{"type":55,"value":1589}," -p",{"type":49,"tag":613,"props":1591,"children":1592},{"style":626},[1593],{"type":55,"value":1594}," \u003C",{"type":49,"tag":613,"props":1596,"children":1597},{"style":631},[1598],{"type":55,"value":1599},"output_di",{"type":49,"tag":613,"props":1601,"children":1603},{"style":1602},"--shiki-light:#90A4AE;--shiki-default:#EEFFFF;--shiki-dark:#BABED8",[1604],{"type":55,"value":1605},"r",{"type":49,"tag":613,"props":1607,"children":1608},{"style":626},[1609],{"type":55,"value":1610},">",{"type":49,"tag":613,"props":1612,"children":1613},{"style":631},[1614],{"type":55,"value":1615},"\u002Fhome",{"type":49,"tag":613,"props":1617,"children":1618},{"style":1602},[1619],{"type":55,"value":1620}," \\\n",{"type":49,"tag":613,"props":1622,"children":1623},{"class":615,"line":637},[1624,1629,1633,1637,1641,1646],{"type":49,"tag":613,"props":1625,"children":1626},{"style":626},[1627],{"type":55,"value":1628},"         \u003C",{"type":49,"tag":613,"props":1630,"children":1631},{"style":631},[1632],{"type":55,"value":1599},{"type":49,"tag":613,"props":1634,"children":1635},{"style":1602},[1636],{"type":55,"value":1605},{"type":49,"tag":613,"props":1638,"children":1639},{"style":626},[1640],{"type":55,"value":1610},{"type":49,"tag":613,"props":1642,"children":1643},{"style":631},[1644],{"type":55,"value":1645},"\u002F.cache\u002Fmatplotlib",{"type":49,"tag":613,"props":1647,"children":1648},{"style":1602},[1649],{"type":55,"value":1620},{"type":49,"tag":613,"props":1651,"children":1652},{"class":615,"line":655},[1653,1657,1661,1665,1669,1674],{"type":49,"tag":613,"props":1654,"children":1655},{"style":626},[1656],{"type":55,"value":1628},{"type":49,"tag":613,"props":1658,"children":1659},{"style":631},[1660],{"type":55,"value":1599},{"type":49,"tag":613,"props":1662,"children":1663},{"style":1602},[1664],{"type":55,"value":1605},{"type":49,"tag":613,"props":1666,"children":1667},{"style":626},[1668],{"type":55,"value":1610},{"type":49,"tag":613,"props":1670,"children":1671},{"style":631},[1672],{"type":55,"value":1673},"\u002F.cache\u002Ftorchinductor",{"type":49,"tag":613,"props":1675,"children":1676},{"style":1602},[1677],{"type":55,"value":1620},{"type":49,"tag":613,"props":1679,"children":1680},{"class":615,"line":683},[1681,1685,1689,1693,1697],{"type":49,"tag":613,"props":1682,"children":1683},{"style":626},[1684],{"type":55,"value":1628},{"type":49,"tag":613,"props":1686,"children":1687},{"style":631},[1688],{"type":55,"value":1599},{"type":49,"tag":613,"props":1690,"children":1691},{"style":1602},[1692],{"type":55,"value":1605},{"type":49,"tag":613,"props":1694,"children":1695},{"style":626},[1696],{"type":55,"value":1610},{"type":49,"tag":613,"props":1698,"children":1699},{"style":631},[1700],{"type":55,"value":1701},"\u002F.cache\u002Fxdg\n",{"type":49,"tag":429,"props":1703,"children":1706},{"className":1704,"code":1705,"language":55},[432],"docker run --gpus 'device=0' --shm-size 16G --ipc=host \\\n  --user \"$(id -u):$(id -g)\" \\\n  -e USER=\"$(id -un)\" \\\n  -e LOGNAME=\"$(id -un)\" \\\n  -e HOME=\u003Coutput_dir>\u002Fhome \\\n  -e MPLCONFIGDIR=\u003Coutput_dir>\u002F.cache\u002Fmatplotlib \\\n  -e TORCHINDUCTOR_CACHE_DIR=\u003Coutput_dir>\u002F.cache\u002Ftorchinductor \\\n  -e XDG_CACHE_HOME=\u003Coutput_dir>\u002F.cache\u002Fxdg \\\n  -v \u003Cdata_root>:\u003Cdata_root>:ro \\\n  -v \u003Coutput_dir>:\u003Coutput_dir> \\\n  \u003Ccontainer> \\\n  depth_net \u003Caction> -e \u003Cspec.yaml>\n",[1707],{"type":49,"tag":69,"props":1708,"children":1709},{"__ignoreMap":437},[1710],{"type":55,"value":1705},{"type":49,"tag":58,"props":1712,"children":1713},{},[1714,1716,1722,1724,1730],{"type":55,"value":1715},"Without ",{"type":49,"tag":69,"props":1717,"children":1719},{"className":1718},[],[1720],{"type":55,"value":1721},"--user \"$(id -u):$(id -g)\"",{"type":55,"value":1723}," the container writes outputs as ",{"type":49,"tag":69,"props":1725,"children":1727},{"className":1726},[],[1728],{"type":55,"value":1729},"nobody:nogroup",{"type":55,"value":1731},", blocking host-side cleanup \u002F retry.",{"type":49,"tag":370,"props":1733,"children":1735},{"id":1734},"step-5-verify",[1736],{"type":55,"value":1737},"Step 5 — Verify",{"type":49,"tag":382,"props":1739,"children":1740},{},[1741,1746,1765,1793,1876],{"type":49,"tag":386,"props":1742,"children":1743},{},[1744],{"type":55,"value":1745},"Container exit code 0",{"type":49,"tag":386,"props":1747,"children":1748},{},[1749,1755,1757,1763],{"type":49,"tag":69,"props":1750,"children":1752},{"className":1751},[],[1753],{"type":55,"value":1754},"status.json",{"type":55,"value":1756}," ",{"type":49,"tag":69,"props":1758,"children":1760},{"className":1759},[],[1761],{"type":55,"value":1762},"kpi",{"type":55,"value":1764}," block populated",{"type":49,"tag":386,"props":1766,"children":1767},{},[1768,1770,1775,1777,1783,1785,1791],{"type":55,"value":1769},"For ",{"type":49,"tag":69,"props":1771,"children":1773},{"className":1772},[],[1774],{"type":55,"value":124},{"type":55,"value":1776},": inspect per-step ",{"type":49,"tag":69,"props":1778,"children":1780},{"className":1779},[],[1781],{"type":55,"value":1782},"train_loss",{"type":55,"value":1784}," directly (the entrypoint reports ",{"type":49,"tag":69,"props":1786,"children":1788},{"className":1787},[],[1789],{"type":55,"value":1790},"Execution status: PASS",{"type":55,"value":1792}," even when loss is NaN)",{"type":49,"tag":386,"props":1794,"children":1795},{},[1796,1797,1802,1804,1810,1811,1817,1818,1824,1825,1831,1832,1838,1839,1845,1847,1853,1854,1860,1861,1867,1869,1875],{"type":55,"value":1769},{"type":49,"tag":69,"props":1798,"children":1800},{"className":1799},[],[1801],{"type":55,"value":132},{"type":55,"value":1803},": rely on ",{"type":49,"tag":69,"props":1805,"children":1807},{"className":1806},[],[1808],{"type":55,"value":1809},"epe",{"type":55,"value":256},{"type":49,"tag":69,"props":1812,"children":1814},{"className":1813},[],[1815],{"type":55,"value":1816},"bp1",{"type":55,"value":256},{"type":49,"tag":69,"props":1819,"children":1821},{"className":1820},[],[1822],{"type":55,"value":1823},"bp2",{"type":55,"value":256},{"type":49,"tag":69,"props":1826,"children":1828},{"className":1827},[],[1829],{"type":55,"value":1830},"bp3",{"type":55,"value":256},{"type":49,"tag":69,"props":1833,"children":1835},{"className":1834},[],[1836],{"type":55,"value":1837},"d1",{"type":55,"value":256},{"type":49,"tag":69,"props":1840,"children":1842},{"className":1841},[],[1843],{"type":55,"value":1844},"rmse",{"type":55,"value":1846}," (the evaluator also emits ",{"type":49,"tag":69,"props":1848,"children":1850},{"className":1849},[],[1851],{"type":55,"value":1852},"abs_rel",{"type":55,"value":256},{"type":49,"tag":69,"props":1855,"children":1857},{"className":1856},[],[1858],{"type":55,"value":1859},"sq_rel",{"type":55,"value":256},{"type":49,"tag":69,"props":1862,"children":1864},{"className":1863},[],[1865],{"type":55,"value":1866},"rmse_log",{"type":55,"value":1868}," which are non-meaningful for stereo — see ",{"type":49,"tag":69,"props":1870,"children":1872},{"className":1871},[],[1873],{"type":55,"value":1874},"references\u002Fparameters-foundation-stereo.md",{"type":55,"value":1338},{"type":49,"tag":386,"props":1877,"children":1878},{},[1879,1880,1885,1887],{"type":55,"value":1769},{"type":49,"tag":69,"props":1881,"children":1883},{"className":1882},[],[1884],{"type":55,"value":139},{"type":55,"value":1886},": artifacts under ",{"type":49,"tag":69,"props":1888,"children":1890},{"className":1889},[],[1891],{"type":55,"value":623},{"type":49,"tag":58,"props":1893,"children":1894},{},[1895,1896,1901,1902,1907,1908,1913,1914,1919,1921,1927,1929,1935],{"type":55,"value":176},{"type":49,"tag":69,"props":1897,"children":1899},{"className":1898},[],[1900],{"type":55,"value":169},{"type":55,"value":183},{"type":49,"tag":69,"props":1903,"children":1905},{"className":1904},[],[1906],{"type":55,"value":132},{"type":55,"value":190},{"type":49,"tag":69,"props":1909,"children":1911},{"className":1910},[],[1912],{"type":55,"value":139},{"type":55,"value":197},{"type":49,"tag":69,"props":1915,"children":1917},{"className":1916},[],[1918],{"type":55,"value":203},{"type":55,"value":1920}," first. Deploy spec templates live in this skill's ",{"type":49,"tag":69,"props":1922,"children":1924},{"className":1923},[],[1925],{"type":55,"value":1926},"references\u002F",{"type":55,"value":1928}," folder with the ",{"type":49,"tag":69,"props":1930,"children":1932},{"className":1931},[],[1933],{"type":55,"value":1934},"spec_template_deploy_*.yaml",{"type":55,"value":1936}," prefix.",{"type":49,"tag":214,"props":1938,"children":1940},{"id":1939},"training-requirements",[1941],{"type":55,"value":1942},"Training Requirements",{"type":49,"tag":382,"props":1944,"children":1945},{},[1946,1956],{"type":49,"tag":386,"props":1947,"children":1948},{},[1949,1954],{"type":49,"tag":390,"props":1950,"children":1951},{},[1952],{"type":55,"value":1953},"Monitoring metric:",{"type":55,"value":1955}," val\u002Floss",{"type":49,"tag":386,"props":1957,"children":1958},{},[1959,1964,1966,1972,1974,1979,1980,1985],{"type":49,"tag":390,"props":1960,"children":1961},{},[1962],{"type":55,"value":1963},"Eval dataset:",{"type":55,"value":1965}," optional. Val dataset configured via ",{"type":49,"tag":69,"props":1967,"children":1969},{"className":1968},[],[1970],{"type":55,"value":1971},"dataset.val_dataset.data_sources",{"type":55,"value":1973}," (each entry needs ",{"type":49,"tag":69,"props":1975,"children":1977},{"className":1976},[],[1978],{"type":55,"value":1517},{"type":55,"value":76},{"type":49,"tag":69,"props":1981,"children":1983},{"className":1982},[],[1984],{"type":55,"value":944},{"type":55,"value":113},{"type":49,"tag":58,"props":1987,"children":1988},{},[1989,1991,1996],{"type":55,"value":1990},"See ",{"type":49,"tag":69,"props":1992,"children":1994},{"className":1993},[],[1995],{"type":55,"value":1316},{"type":55,"value":1997}," for the per-action dataset-requirements table and every action's mandatory data-source override block.",{"type":49,"tag":214,"props":1999,"children":2001},{"id":2000},"parameters-metrics-multi-gpu-exporttrt-hardware",[2002],{"type":55,"value":2003},"Parameters, Metrics, Multi-GPU, Export\u002FTRT, Hardware",{"type":49,"tag":58,"props":2005,"children":2006},{},[2007,2008,2013,2015,2021,2022,2028,2030,2036,2038,2044,2046,2052,2053,2059,2060,2066,2067,2073,2075,2081,2083,2089],{"type":55,"value":1990},{"type":49,"tag":69,"props":2009,"children":2011},{"className":2010},[],[2012],{"type":55,"value":1874},{"type":55,"value":2014}," for the full Important Parameters list (incl. ",{"type":49,"tag":69,"props":2016,"children":2018},{"className":2017},[],[2019],{"type":55,"value":2020},"model.encoder",{"type":55,"value":1756},{"type":49,"tag":69,"props":2023,"children":2025},{"className":2024},[],[2026],{"type":55,"value":2027},"vits",{"type":55,"value":2029}," override, ",{"type":49,"tag":69,"props":2031,"children":2033},{"className":2032},[],[2034],{"type":55,"value":2035},"model.max_disparity",{"type":55,"value":2037}," default 416, ",{"type":49,"tag":69,"props":2039,"children":2041},{"className":2040},[],[2042],{"type":55,"value":2043},"model.volume_dim",{"type":55,"value":2045}," no-op note, ",{"type":49,"tag":69,"props":2047,"children":2049},{"className":2048},[],[2050],{"type":55,"value":2051},"dataset.baseline",{"type":55,"value":126},{"type":49,"tag":69,"props":2054,"children":2056},{"className":2055},[],[2057],{"type":55,"value":2058},"dataset.focal_x",{"type":55,"value":126},{"type":49,"tag":69,"props":2061,"children":2063},{"className":2062},[],[2064],{"type":55,"value":2065},"train.precision",{"type":55,"value":126},{"type":49,"tag":69,"props":2068,"children":2070},{"className":2069},[],[2071],{"type":55,"value":2072},"export.batch_size",{"type":55,"value":2074},"), the Evaluation Metrics table, Multi-GPU \u002F Multi-Node launch keys, Export \u002F TRT Defaults (",{"type":49,"tag":69,"props":2076,"children":2078},{"className":2077},[],[2079],{"type":55,"value":2080},"opset_version",{"type":55,"value":2082},"\u002F",{"type":49,"tag":69,"props":2084,"children":2086},{"className":2085},[],[2087],{"type":55,"value":2088},"on_cpu",{"type":55,"value":2090}," pairing, NGC 576×960 settings), and Hardware requirements.",{"type":49,"tag":214,"props":2092,"children":2094},{"id":2093},"error-patterns-and-troubleshooting",[2095],{"type":55,"value":2096},"Error Patterns and Troubleshooting",{"type":49,"tag":58,"props":2098,"children":2099},{},[2100,2101,2106,2108,2114,2115,2120,2122,2128,2130,2135,2137,2142],{"type":55,"value":1990},{"type":49,"tag":69,"props":2102,"children":2104},{"className":2103},[],[2105],{"type":55,"value":1449},{"type":55,"value":2107}," for disparity overflow, smoke-test shape mismatch, missing pretrained paths, the ",{"type":49,"tag":69,"props":2109,"children":2111},{"className":2110},[],[2112],{"type":55,"value":2113},"encoder",{"type":55,"value":256},{"type":49,"tag":69,"props":2116,"children":2118},{"className":2117},[],[2119],{"type":55,"value":944},{"type":55,"value":2121}," struct errors, the ",{"type":49,"tag":69,"props":2123,"children":2125},{"className":2124},[],[2126],{"type":55,"value":2127},"depth_net_stereo: not found",{"type":55,"value":2129}," entrypoint note, the pyt-vs-deploy ",{"type":49,"tag":69,"props":2131,"children":2133},{"className":2132},[],[2134],{"type":55,"value":1397},{"type":55,"value":2136}," discussion, and the deploy ",{"type":49,"tag":69,"props":2138,"children":2140},{"className":2139},[],[2141],{"type":55,"value":132},{"type":55,"value":2143}," scalar-conversion failure.",{"type":49,"tag":214,"props":2145,"children":2147},{"id":2146},"spec-param-parent-model-inference",[2148],{"type":55,"value":2149},"Spec Param \u002F Parent Model Inference",{"type":49,"tag":58,"props":2151,"children":2152},{},[2153,2154,2160,2162,2168,2169,2175,2177,2182,2184,2190,2192,2198,2199,2205],{"type":55,"value":1990},{"type":49,"tag":69,"props":2155,"children":2157},{"className":2156},[],[2158],{"type":55,"value":2159},"references\u002Fcheckpoint-inference-mappings-foundation-stereo.md",{"type":55,"value":2161}," for the checkpoint-resolution rules (",{"type":49,"tag":69,"props":2163,"children":2165},{"className":2164},[],[2166],{"type":55,"value":2167},"model_epoch_\u003Cepoch>_step_\u003Cstep>.pth",{"type":55,"value":126},{"type":49,"tag":69,"props":2170,"children":2172},{"className":2171},[],[2173],{"type":55,"value":2174},"dn_model_latest.pth",{"type":55,"value":2176}," policy), the absence of parent PyT ",{"type":49,"tag":69,"props":2178,"children":2180},{"className":2179},[],[2181],{"type":55,"value":169},{"type":55,"value":2183},", and the full per-action inference-mapping table from ",{"type":49,"tag":69,"props":2185,"children":2187},{"className":2186},[],[2188],{"type":55,"value":2189},"depth_net_stereo.config.json",{"type":55,"value":2191}," (including ",{"type":49,"tag":69,"props":2193,"children":2195},{"className":2194},[],[2196],{"type":55,"value":2197},"parent_model",{"type":55,"value":256},{"type":49,"tag":69,"props":2200,"children":2202},{"className":2201},[],[2203],{"type":55,"value":2204},"parent_job_id",{"type":55,"value":2206}," resolution).",{"type":49,"tag":214,"props":2208,"children":2210},{"id":2209},"deployment",[2211],{"type":55,"value":2212},"Deployment",{"type":49,"tag":382,"props":2214,"children":2215},{},[2216],{"type":49,"tag":386,"props":2217,"children":2218},{},[2219],{"type":49,"tag":2220,"props":2221,"children":2222},"a",{"href":203},[2223],{"type":55,"value":2224},"tao-deploy-foundation-stereo",{"type":49,"tag":2226,"props":2227,"children":2228},"style",{},[2229],{"type":55,"value":2230},"html .light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html.light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"items":2232,"total":2389},[2233,2251,2267,2278,2290,2304,2317,2331,2344,2355,2369,2378],{"slug":2234,"name":2234,"fn":2235,"description":2236,"org":2237,"tags":2238,"stars":2248,"repoUrl":2249,"updatedAt":2250},"nemoclaw-user-guide","retrieve NemoClaw documentation and configuration","Guides human users' AI agents to the NemoClaw docs MCP server and canonical Fern documentation in Markdown form. Use when users ask how to install, configure, operate, troubleshoot, secure, or learn NemoClaw with an AI coding assistant. Trigger keywords - nemoclaw docs, use nemoclaw with ai agent, nemoclaw mcp docs, nemoclaw install help, nemoclaw quickstart, nemoclaw markdown docs, llms.txt, agent skills.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2239,2242,2245],{"name":2240,"slug":2241,"type":15},"Documentation","documentation",{"name":2243,"slug":2244,"type":15},"MCP","mcp",{"name":2246,"slug":2247,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":2252,"name":2252,"fn":2253,"description":2254,"org":2255,"tags":2256,"stars":2264,"repoUrl":2265,"updatedAt":2266},"mcore-build-and-dependency","manage Megatron-LM development environments","Container-based dev environment setup and dependency management for Megatron-LM. Covers acquiring and launching the CI container, uv package management, and updating uv.lock.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2257,2260,2261],{"name":2258,"slug":2259,"type":15},"Containers","containers",{"name":2212,"slug":2209,"type":15},{"name":2262,"slug":2263,"type":15},"Python","python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":2268,"name":2268,"fn":2269,"description":2270,"org":2271,"tags":2272,"stars":2264,"repoUrl":2265,"updatedAt":2277},"mcore-bump-base-image","update NVIDIA PyTorch base images","Bump the NVIDIA PyTorch base image (`nvcr.io\u002Fnvidia\u002Fpytorch:YY.MM-py3`) used by Megatron-LM CI. Covers the two pin sites (GitHub CI in `docker\u002F.ngc_version.dev` and GitLab CI in `.gitlab\u002Fstages\u002F01.build.yml`), the post-bump CI loop (re-run functional tests, refresh golden values, mark broken tests), and the gotchas that bit PRs",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2273,2276],{"name":2274,"slug":2275,"type":15},"CI\u002FCD","ci-cd",{"name":2212,"slug":2209,"type":15},"2026-07-14T05:25:59.97109",{"slug":2279,"name":2279,"fn":2280,"description":2281,"org":2282,"tags":2283,"stars":2264,"repoUrl":2265,"updatedAt":2289},"mcore-cicd","manage CI\u002FCD pipelines for Megatron-LM","CI\u002FCD reference for Megatron-LM. Covers CI pipeline structure, PR scope labels, triggering internal GitLab CI (which force-pushes the current branch to a pull-request\u002FBRANCH ref — always dry-run and verify the destination first; never run against shared or protected branches), and CI failure investigation.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2284,2285,2286],{"name":2274,"slug":2275,"type":15},{"name":2212,"slug":2209,"type":15},{"name":2287,"slug":2288,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":2291,"name":2291,"fn":2292,"description":2293,"org":2294,"tags":2295,"stars":2264,"repoUrl":2265,"updatedAt":2303},"mcore-create-issue","investigate CI failures and create issues","Investigate a failing GitHub Actions run or job and create a GitHub issue for the failure.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2296,2299,2300],{"name":2297,"slug":2298,"type":15},"Debugging","debugging",{"name":2287,"slug":2288,"type":15},{"name":2301,"slug":2302,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":2305,"name":2305,"fn":2306,"description":2307,"org":2308,"tags":2309,"stars":2264,"repoUrl":2265,"updatedAt":2316},"mcore-linting-and-formatting","lint and format Megatron-LM code","Linting and formatting for Megatron-LM. Covers running autoformat.sh, tools (ruff, black, isort, pylint, mypy), and code style rules.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2310,2313],{"name":2311,"slug":2312,"type":15},"Best Practices","best-practices",{"name":2314,"slug":2315,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":2318,"name":2318,"fn":2319,"description":2320,"org":2321,"tags":2322,"stars":2264,"repoUrl":2265,"updatedAt":2330},"mcore-migrate-gpt-to-hybrid","migrate Megatron-LM models to HybridModel","Migration guide for moving Megatron Core GPTModel checkpoints, model providers, training commands, and layer mappings to HybridModel.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2323,2326,2329],{"name":2324,"slug":2325,"type":15},"Machine Learning","machine-learning",{"name":2327,"slug":2328,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":2332,"name":2332,"fn":2333,"description":2334,"org":2335,"tags":2336,"stars":2264,"repoUrl":2265,"updatedAt":2343},"mcore-onboard-gb200-1node-tests","onboard functional tests for GB200","Onboard 1-node GitHub MR functional tests for GB200 from existing mr-scoped 2-node tests.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2337,2340],{"name":2338,"slug":2339,"type":15},"QA","qa",{"name":2341,"slug":2342,"type":15},"Testing","testing","2026-07-14T05:25:53.673039",{"slug":2345,"name":2345,"fn":2346,"description":2347,"org":2348,"tags":2349,"stars":2264,"repoUrl":2265,"updatedAt":2354},"mcore-run-on-slurm","launch distributed training jobs on SLURM","How to launch distributed Megatron-LM training jobs on a SLURM cluster. Covers a minimal sbatch skeleton, environment-variable setup for torch.distributed.run, CUDA_DEVICE_MAX_CONNECTIONS rules across hardware and parallelism modes, container conventions, monitoring, and per-rank failure diagnosis.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2350,2351],{"name":2212,"slug":2209,"type":15},{"name":2352,"slug":2353,"type":15},"Infrastructure","infrastructure","2026-07-14T05:25:49.362534",{"slug":2356,"name":2356,"fn":2357,"description":2358,"org":2359,"tags":2360,"stars":2264,"repoUrl":2265,"updatedAt":2368},"mcore-split-pr","split pull requests to reduce review load","Split a PR into multiple PRs to reduce the number of required CODEOWNERS reviewer groups.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2361,2364,2365],{"name":2362,"slug":2363,"type":15},"Code Review","code-review",{"name":2287,"slug":2288,"type":15},{"name":2366,"slug":2367,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":2370,"name":2370,"fn":2371,"description":2372,"org":2373,"tags":2374,"stars":2264,"repoUrl":2265,"updatedAt":2377},"mcore-testing","run and manage Megatron-LM tests","Test system for Megatron-LM. Covers test layout, recipe YAML structure, adding and running unit and functional tests, golden values, marker filters, and CI parity.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2375,2376],{"name":2338,"slug":2339,"type":15},{"name":2341,"slug":2342,"type":15},"2026-07-14T05:25:54.928983",{"slug":2379,"name":2379,"fn":2380,"description":2381,"org":2382,"tags":2383,"stars":2264,"repoUrl":2265,"updatedAt":2388},"nightly-sync","manage nightly main-to-dev sync workflows","Domain knowledge for the nightly main-to-dev sync workflow. Covers merge strategy, CI architecture, failure investigation, and known issues.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2384,2387],{"name":2385,"slug":2386,"type":15},"Automation","automation",{"name":2274,"slug":2275,"type":15},"2026-07-30T05:29:03.275638",496,{"items":2391,"total":2485},[2392,2409,2419,2433,2443,2456,2471],{"slug":2393,"name":2393,"fn":2394,"description":2395,"org":2396,"tags":2397,"stars":23,"repoUrl":24,"updatedAt":2408},"accelerated-computing-cudf","accelerate data processing with cuDF","Official NVIDIA-authored guidance for NVIDIA cuDF GPU DataFrames, pandas acceleration, dask-cuDF, ETL, joins, groupby, CSV\u002FParquet I\u002FO, nullable semantics, and multi-GPU DataFrame workloads.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2398,2401,2404,2405],{"name":2399,"slug":2400,"type":15},"Data Analysis","data-analysis",{"name":2402,"slug":2403,"type":15},"Data Engineering","data-engineering",{"name":9,"slug":8,"type":15},{"name":2406,"slug":2407,"type":15},"Performance","performance","2026-07-14T05:28:43.176466",{"slug":2410,"name":2410,"fn":2411,"description":2412,"org":2413,"tags":2414,"stars":23,"repoUrl":24,"updatedAt":2418},"aiq-deploy","deploy and manage NVIDIA AI-Q infrastructure","Use when asked to install, deploy, run, validate, troubleshoot, or stop NVIDIA AI-Q Blueprint infrastructure.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2415,2416,2417],{"name":2212,"slug":2209,"type":15},{"name":2352,"slug":2353,"type":15},{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":2420,"name":2420,"fn":2421,"description":2422,"org":2423,"tags":2424,"stars":23,"repoUrl":24,"updatedAt":2432},"aiq-research","conduct deep research with AI-Q","Use when asked to run deep research or AI-Q research through a reachable NVIDIA AI-Q Blueprint backend.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2425,2428,2429],{"name":2426,"slug":2427,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":2430,"slug":2431,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":2434,"name":2434,"fn":2435,"description":2436,"org":2437,"tags":2438,"stars":23,"repoUrl":24,"updatedAt":2442},"amc-run-sample-calibration","run AMC sample dataset calibration","Run end-to-end calibration on the shipped sample dataset (sdg_08_2_sample_data_010926.zip) against a running AMC microservice. Use when user says 'test sample dataset', 'run sample calibration', 'verify AMC install', or 'launch and test'.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2439,2440,2441],{"name":2399,"slug":2400,"type":15},{"name":9,"slug":8,"type":15},{"name":2341,"slug":2342,"type":15},"2026-07-17T05:29:03.913266",{"slug":2444,"name":2444,"fn":2445,"description":2446,"org":2447,"tags":2448,"stars":23,"repoUrl":24,"updatedAt":2455},"amc-run-video-calibration","calibrate video datasets with AutoMagicCalib","Calibrate a new dataset from pre-recorded video files via the AutoMagicCalib REST API. Use when user has local MP4s and says 'calibrate my videos', 'run AMC on these videos', or similar. For RTSP\u002Flive streams, use amc-run-rtsp-calibration instead.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2449,2450,2451,2452],{"name":2385,"slug":2386,"type":15},{"name":18,"slug":19,"type":15},{"name":9,"slug":8,"type":15},{"name":2453,"slug":2454,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":2457,"name":2457,"fn":2458,"description":2459,"org":2460,"tags":2461,"stars":23,"repoUrl":24,"updatedAt":2470},"amc-setup-calibration-stack","deploy AutoMagicCalib microservice with Docker","Launch AutoMagicCalib microservice and web UI from NGC release images via Docker Compose. Use when user says 'deploy auto calibration', 'launch auto calibration', 'launch AMC', 'start MS+UI', or 'set up auto-magic-calib'. Requires NGC API key.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2462,2463,2466,2467],{"name":2212,"slug":2209,"type":15},{"name":2464,"slug":2465,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":2468,"slug":2469,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":2472,"name":2472,"fn":2473,"description":2474,"org":2475,"tags":2476,"stars":23,"repoUrl":24,"updatedAt":2484},"cudaq-guide","develop quantum applications with CUDA-Q","CUDA-Q onboarding guide for installation, test programs, GPU simulation, QPU hardware, and quantum applications.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2477,2478,2481],{"name":9,"slug":8,"type":15},{"name":2479,"slug":2480,"type":15},"Quantum Computing","quantum-computing",{"name":2482,"slug":2483,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305]