tao-train-foundation-stereo

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Depth Net Stereo

Stereo depth estimation using FoundationStereo architecture. Predicts disparity maps from stereo image pairs for 3D reconstruction.

Uses pretrained Depth Anything v2 and EdgeNeXt encoders. Set

model.stereo_backbone.depth_anything_v2_pretrained_path

and

model.stereo_backbone.edgenext_pretrained_path

The mono and stereo skills both invoke the unified TAO

depth_net

CLI inside the container; the mono/stereo family is selected via

model.model_type

(e.g.,

FoundationStereo

For TAO Deploy TensorRT actions (

gen_trt_engine

, TensorRT

evaluate

, and TensorRT

inference

), read

references/tao-deploy-foundation-stereo.md

first. The deploy spec template lives in this skill's

references/spec_template_deploy.yaml

使用FoundationStereo架构进行立体深度估计。从立体图像对中预测视差图，用于3D重建。

采用预训练的Depth Anything v2和EdgeNeXt编码器。需设置

model.stereo_backbone.depth_anything_v2_pretrained_path

和

model.stereo_backbone.edgenext_pretrained_path

。

单目和立体技能都会调用容器内统一的TAO

depth_net

CLI；通过

model.model_type

选择单目/立体系列（例如

FoundationStereo

）。

对于TAO Deploy TensorRT操作（

gen_trt_engine

、TensorRT

evaluate

和TensorRT

inference

），请先阅读

references/tao-deploy-foundation-stereo.md

。部署规范模板位于本技能的

references/spec_template_deploy.yaml

中。

Train Action Policy

训练操作策略

This model is AutoML-enabled at the model layer. Before handling any train-stage request, read

references/skill_info.yaml

and resolve the run override from either an explicit

automl_policy

value or the user's workflow request. Treat phrases like "turn off AutoML", "disable AutoML", "no HPO", or "plain training" as

automl_policy: off

for this run only; otherwise default to

auto

. When

automl_policy: auto

automl_enabled: true

, and both

schemas/train.schema.json

and

references/spec_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/application overrides for datasets, specs, output directories, GPU/platform settings, parent checkpoints, and

automl_policy

. Use direct model training only when

automl_policy: off

or the packaged train schema/template is missing; in the missing-schema case, report that AutoML is enabled but not runnable for this model until schemas are generated.

Non-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.

该模型在模型层支持AutoML。处理任何训练阶段的请求前，请阅读

references/skill_info.yaml

，并通过显式的

automl_policy

值或用户的工作流请求解析运行覆盖配置。将“turn off AutoML”、“disable AutoML”、“no HPO”或“plain training”这类表述视为本次运行的

automl_policy: off

；否则默认设为

auto

。当

automl_policy: auto

、

automl_enabled: true

，且已打包

schemas/train.schema.json

和

references/spec_template_train.yaml

时，默认将训练操作通过

tao-skill-bank:tao-run-automl

路由，并传入该模型的

skill_dir

。保留数据集、规范、输出目录、GPU/平台设置、父检查点和

automl_policy

的工作流/应用覆盖配置。仅当

automl_policy: off

或打包的训练架构/模板缺失时，才使用直接模型训练；在架构缺失的情况下，需报告AutoML已启用，但在生成架构前无法针对该模型运行。

非训练操作（如

evaluate

、

inference

、

export

和部署流程）仍在本模型技能中执行。每次运行的

automl_policy

覆盖配置不会更改模型元数据。

Workflow

工作流

Prerequisites — data accessibility

前提条件——数据可访问性

Your dataset (left + right images + GT disparity) must be reachable from inside the container:

SDK runner: place files at the S3 paths the runner resolves (the
```
S3_TRAIN
```
/
```
S3_EVAL
```
placeholders shown in Typical Spec Overrides). The runner handles S3 → container-path mounting transparently.
Direct
docker run
(e.g. local testing): mount the host dataset root read-only at the same in-container path:

docker run ... -v <host_data_root>:<host_data_root>:ro <container> ...

The same accessibility requirement applies to the

<output_dir>

written by all actions.

你的数据集（左图+右图+GT视差）必须能从容器内部访问：

SDK运行器：将文件放置在运行器解析的S3路径下（即典型规范覆盖中显示的
```
S3_TRAIN
```
/
```
S3_EVAL
```
占位符）。运行器会自动处理S3到容器路径的挂载。
直接
docker run
（例如本地测试）：将主机数据集根目录以只读方式挂载到容器内相同路径：

docker run ... -v <host_data_root>:<host_data_root>:ro <container> ...

所有操作写入的

<output_dir>

也需满足相同的可访问性要求。

Step 1 — Annotation file

步骤1——标注文件

Per-line annotation file referenced by

data_sources[*].data_file

Columns	Format	Use
2	`<left> <right>`	Stereo inference (no GT)
3	`<left> <right> <disparity>`	Stereo with GT
4	`<left> <right> <disparity> <occlusion_mask>`	Stereo with GT and occlusion mask

If you already have one, point to it. Otherwise generate via

depth_net convert

depth_net convert -e <convert_spec.yaml>

convert_spec.yaml

template (stereo):

yaml

data_root: <directory 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>
image_dir_pattern: [<substring matching left image paths>]
right_dir_pattern: [<substring matching right image paths>]
depth_dir_pattern: [<substring matching GT disparity paths>]
nocc_dir_pattern: []                 # optional, occlusion mask paths
image_extension: '.png'  # always include the leading dot
depth_extension: '.png'  # form must match image_extension (the swap is a substring replace)
nocc_extension: ''
split_ratio: 0.0        # 0.0/1.0 = test-only; 0.8 = 80/20 train+val

convert

walks

data_root

recursively, selects paths whose path-string contains all substrings in

image_dir_pattern

(AND-filter), then derives right / depth / 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

im1

disp0GT

for Middlebury).

data_sources[*].data_file

引用的每行标注文件：

列数	格式	用途
2	`<left> <right>`	立体推理（无GT）
3	`<left> <right> <disparity>`	带GT的立体任务
4	`<left> <right> <disparity> <occlusion_mask>`	带GT和遮挡掩码的立体任务

如果已有标注文件，直接指向它。否则通过

depth_net convert

生成：

depth_net convert -e <convert_spec.yaml>

convert_spec.yaml

模板（立体任务）：

yaml

data_root: <目录，其子目录为包含图像+深度文件的场景文件夹；convert会递归遍历data_root，但要求场景子目录位于data_root下一级>
image_dir_pattern: [<匹配左图路径的子字符串>]
right_dir_pattern: [<匹配右图路径的子字符串>]
depth_dir_pattern: [<匹配GT视差路径的子字符串>]
nocc_dir_pattern: []                 # 可选，遮挡掩码路径
image_extension: '.png'  # 必须包含前导点
depth_extension: '.png'  # 格式必须与image_extension匹配（通过子字符串替换实现转换）
nocc_extension: ''
split_ratio: 0.0        # 0.0/1.0 = 仅测试集；0.8 = 80/20 训练+验证集

convert

会递归遍历

data_root

，选择路径字符串包含

image_dir_pattern

中所有子字符串的路径（与过滤），然后通过将

image_dir_pattern[0]

替换为对应模式的第一个元素并交换扩展名，推导右图/深度/掩码路径。检查你的数据集目录结构，区分左图、右图和GT的子字符串（例如Middlebury数据集的

im0

im1

disp0GT

）。

Step 2 — Pair

model_type

and

dataset_name

based on your data

步骤2——根据数据匹配

model_type

和

dataset_name

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

GenericDataset

only for layouts that do not match any registered class.

Data category	`model_type`	`dataset_name`
Middlebury data	`FoundationStereo`	`Middlebury`
KITTI data	`FoundationStereo`	`Kitti`
ETH3D data	`FoundationStereo`	`Eth3d`
FSD synthetic data	`FoundationStereo`	`FSD`
IsaacReal synthetic data	`FoundationStereo`	`IsaacRealDataset`
Crestereo synthetic data	`FoundationStereo`	`Crestereo`
Other / non-canonical layout	`FoundationStereo`	`GenericDataset`

See Training Requirements → Formats for the full registered-class list. The 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/tao-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

Kitti

Eth3d

FSD

IsaacRealDataset

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.

当你的数据布局与支持的数据集匹配时，优先使用特定数据集类——它会应用类特定的路径约定、评估裁剪，以及（如适用）遮挡掩码处理。仅当布局不匹配任何已注册类时，才回退到

GenericDataset

。

数据类别	`model_type`	`dataset_name`
Middlebury数据	`FoundationStereo`	`Middlebury`
KITTI数据	`FoundationStereo`	`Kitti`
ETH3D数据	`FoundationStereo`	`Eth3d`
FSD合成数据	`FoundationStereo`	`FSD`
IsaacReal合成数据	`FoundationStereo`	`IsaacRealDataset`
Crestereo合成数据	`FoundationStereo`	`Crestereo`
其他/非标准布局	`FoundationStereo`	`GenericDataset`

查看训练要求→格式获取完整的已注册类列表。同一个

dataset_name

值适用于训练和评估操作（这些操作均使用带GT视差的3列或4列标注）。部署端的

evaluate

操作遵循相同规则——请参阅

references/tao-deploy-foundation-stereo.md

。对于使用2列标注（左图+右图，无GT）的推理，无论数据布局如何，都使用

dataset_name: GenericDataset

——特定数据集类（

Middlebury

Kitti

Eth3d

FSD

IsaacRealDataset

Crestereo

）要求3列输入，在数据加载器层面会拒绝2列标注。对于使用3列标注（左图+右图+GT）的推理，特定数据集类是适用的。

Step 3 — Write spec yaml from Typical Spec Overrides

步骤3——根据典型规范覆盖编写spec yaml

Copy the action block from

references/foundation-stereo-spec-overrides.md

(per-action

spec_overrides

, mandatory data sources). Replace:

```
model.model_type
```
from Step 2 (typically
```
FoundationStereo
```
)

dataset.<...>.data_sources[*].dataset_name

from Step 2

```
dataset.<...>.data_sources[*].data_file
```
with the path from Step 1

For deploy-side

evaluate

: enforce

dataset.test_dataset.batch_size: 1

(see

references/tao-deploy-foundation-stereo.md

Shape consistency: the

crop_size

dataset.test_dataset.augmentation.crop_size

should match

export.input_height

input_width

so the trained-model evaluator and the deploy-side TensorRT evaluator operate at the same shape — see

references/foundation-stereo-troubleshooting.md

从

references/foundation-stereo-spec-overrides.md

复制操作块（每个操作的

spec_overrides

，必填数据源）。替换：

步骤2中的
```
model.model_type
```
（通常为
```
FoundationStereo
```
）

步骤2中的

dataset.<...>.data_sources[*].dataset_name

步骤1中的路径替换
```
dataset.<...>.data_sources[*].data_file
```

对于部署端的

evaluate

：强制设置

dataset.test_dataset.batch_size: 1

（请参阅

references/tao-deploy-foundation-stereo.md

）。

形状一致性：

dataset.test_dataset.augmentation.crop_size

中的

crop_size

应与

export.input_height

input_width

匹配，以便训练模型评估器和部署端TensorRT评估器在相同形状下运行——请参阅

references/foundation-stereo-troubleshooting.md

。

Step 4 — Run

步骤4——运行

docker run --gpus 'device=0' --shm-size 16G --ipc=host \
  --user $(id -u):$(id -g) \
  -v <data_root>:<data_root>:ro \
  -v <output_dir>:<output_dir> \
  <container> \
  depth_net <action> -e <spec.yaml>

Without

--user $(id -u):$(id -g)

the container writes outputs as

nobody:nogroup

, blocking host-side cleanup / retry.

docker run --gpus 'device=0' --shm-size 16G --ipc=host \
  --user $(id -u):$(id -g) \
  -v <data_root>:<data_root>:ro \
  -v <output_dir>:<output_dir> \
  <container> \
  depth_net <action> -e <spec.yaml>

如果不添加

--user $(id -u):$(id -g)

，容器会以

nobody:nogroup

身份写入输出，导致主机端无法清理/重试。

Step 5 — Verify

步骤5——验证

Container exit code 0
```
status.json
```
```
kpi
```
block populated
For
```
train
```
: inspect per-step
```
train_loss
```
directly (the entrypoint reports
```
Execution status: PASS
```
even when loss is NaN)

For

evaluate

: rely on

epe

bp1

bp2

bp3

d1

rmse

(the evaluator also emits

abs_rel

sq_rel

rmse_log

which are non-meaningful for stereo — see

references/foundation-stereo-parameters.md

)

For
```
inference
```
: artifacts under
```
results_dir
```

For TAO Deploy TensorRT actions (

gen_trt_engine

, TensorRT

evaluate

, and TensorRT

inference

), read

references/tao-deploy-foundation-stereo.md

first. Deploy spec templates live in this skill's

references/

folder with the

spec_template_deploy_*.yaml

prefix.

容器退出码为0
```
status.json
```
的
```
kpi
```
块已填充
对于
```
train
```
：直接检查每一步的
```
train_loss
```
（即使损失为NaN，入口点仍会报告
```
Execution status: PASS
```
）
对于
```
evaluate
```
：参考
```
epe
```
/
```
bp1
```
/
```
bp2
```
/
```
bp3
```
/
```
d1
```
/
```
rmse
```
（评估器还会输出
```
abs_rel
```
/
```
sq_rel
```
/
```
rmse_log
```
，但这些在立体任务中无意义——请参阅
```
references/foundation-stereo-parameters.md
```
）
对于
```
inference
```
：
```
results_dir
```
下存在输出产物

对于TAO Deploy TensorRT操作（

gen_trt_engine

、TensorRT

evaluate

和TensorRT

inference

），请先阅读

references/tao-deploy-foundation-stereo.md

。部署规范模板位于本技能的

references/

文件夹中，前缀为

spec_template_deploy_*.yaml

。

Training Requirements

训练要求

Valid
dataset_name
values for stereo
data_sources
(case-insensitive):

FSD

IsaacRealDataset

Crestereo

Middlebury

Eth3d

Kitti

GenericDataset

Monitoring metric: val/loss

立体

data_sources

的有效

dataset_name

值（大小写不敏感）：

FSD

、

IsaacRealDataset

、

Crestereo

、

Middlebury

、

Eth3d

、

Kitti

、

GenericDataset

监控指标：val/loss

Per-Action Dataset Requirements

各操作的数据集要求

Action	Spec Key	Source	Files	List?
evaluate	dataset.test_dataset.data_sources	eval_dataset	data_file: annotations.txt + dataset_name	Yes
inference	dataset.infer_dataset.data_sources	inference_dataset	data_file: annotations.txt + dataset_name	Yes
quantize	dataset.train_dataset.data_sources	train_datasets	data_file: annotations.txt + dataset_name	Yes
quantize	dataset.val_dataset.data_sources	eval_dataset	data_file: annotations.txt + dataset_name	Yes
quantize	dataset.quant_calibration_dataset.images_dir	train_datasets	images.tar.gz	No
train	dataset.train_dataset.data_sources	train_datasets	data_file: annotations.txt + dataset_name	Yes
train	dataset.val_dataset.data_sources	eval_dataset	data_file: annotations.txt + dataset_name	Yes

操作	规范键	来源	文件	是否为列表？
evaluate	dataset.test_dataset.data_sources	eval_dataset	data_file: annotations.txt + dataset_name	是
inference	dataset.infer_dataset.data_sources	inference_dataset	data_file: annotations.txt + dataset_name	是
quantize	dataset.train_dataset.data_sources	train_datasets	data_file: annotations.txt + dataset_name	是
quantize	dataset.val_dataset.data_sources	eval_dataset	data_file: annotations.txt + dataset_name	是
quantize	dataset.quant_calibration_dataset.images_dir	train_datasets	images.tar.gz	否
train	dataset.train_dataset.data_sources	train_datasets	data_file: annotations.txt + dataset_name	是
train	dataset.val_dataset.data_sources	eval_dataset	data_file: annotations.txt + dataset_name	是

Typical Spec Overrides

典型规范覆盖

Data source overrides are mandatory for every action — the agent MUST construct data source paths from the Per-Action Dataset Requirements table above and include them in

spec_overrides

. Each

data_sources

entry is a dict with two mandatory fields:

data_file

and

dataset_name

See

references/foundation-stereo-spec-overrides.md

for the full per-action

spec_overrides

blocks (train, evaluate, export, gen_trt_engine, inference, quantize) with

S3_TRAIN

S3_EVAL

placeholders.

数据源覆盖对每个操作都是必填项——代理必须根据上述各操作数据集要求表构建数据源路径，并将其包含在

spec_overrides

中。每个

data_sources

条目是一个字典，包含两个必填字段：

data_file

和

dataset_name

。

完整的各操作

spec_overrides

块（train、evaluate、export、gen_trt_engine、inference、quantize）及

S3_TRAIN

S3_EVAL

占位符，请参阅

references/foundation-stereo-spec-overrides.md

。

Eval Dataset

评估数据集

Optional. Val dataset configured via

dataset.val_dataset.data_sources

(each entry needs

data_file

and

dataset_name

可选。验证数据集通过

dataset.val_dataset.data_sources

配置（每个条目需要

data_file

和

dataset_name

）。

Important Parameters

重要参数

Key defaults:

model.model_type

FoundationStereo

(only selectable type);

model.encoder

(top-level, not under

stereo_backbone

) schema default

vitl

but FS small NGC ckpt requires
vits
, override explicitly;

model.max_disparity

default 416;

train.optim.lr

default 1e-4;

train.precision

fp32 (recommended) or fp16 (no bf16);

export.batch_size

default

-1

. The

workers

field name is

workers

, not

num_workers

See

references/foundation-stereo-parameters.md

for the full parameter glossary (all

model.*

dataset.*

train.*

export.*

fields with defaults and ranges) and the Evaluation Metrics reference (which

epe

bp*

d1

rmse

to trust and why

abs_rel

sq_rel

rmse_log

are non-meaningful for stereo).

关键默认值：

model.model_type

FoundationStereo

（唯一可选类型）；

model.encoder

（顶层，不在

stereo_backbone

下）架构默认值为

vitl

，但FS小型NGC检查点需要显式覆盖为
vits
；

model.max_disparity

默认值为416；

train.optim.lr

默认值为1e-4；

train.precision

为fp32（推荐）或fp16（不支持bf16）；

export.batch_size

默认值为

-1

。

workers

字段名称为

workers

，而非

num_workers

。

完整的参数术语表（所有

model.*

、

dataset.*

、

train.*

、

export.*

字段及其默认值和范围）以及评估指标参考（应信任哪些

epe

bp*

d1

rmse

，以及为何

abs_rel

sq_rel

rmse_log

在立体任务中无意义），请参阅

references/foundation-stereo-parameters.md

。

Multi-GPU / Multi-Node

多GPU/多节点

Launch method: Lightning-managed (single

python

process, Lightning spawns workers).

Spec Key	Description	Default
`train.num_gpus`	Number of GPUs	1
`train.gpu_ids`	GPU device indices	[0]
`train.num_nodes`	Number of nodes	1
`train.distributed_strategy`	`ddp` or `fsdp`	`ddp`

Same DDP/FSDP behavior as depth-net-mono. Multi-node requires

WORLD_SIZE

NODE_RANK

MASTER_ADDR

MASTER_PORT

env vars.

启动方式：Lightning管理（单个

python

进程，Lightning生成工作进程）。

规范键	描述	默认值
`train.num_gpus`	GPU数量	1
`train.gpu_ids`	GPU设备索引	[0]
`train.num_nodes`	节点数量	1
`train.distributed_strategy`	`ddp` 或 `fsdp`	`ddp`

与depth-net-mono具有相同的DDP/FSDP行为。多节点需要

WORLD_SIZE

、

NODE_RANK

、

MASTER_ADDR

、

MASTER_PORT

环境变量。

Export / TRT Defaults

导出/TRT默认值

TRT data types FP32 / FP16. Static-shape ONNX (

export.batch_size: 1

) and batch-only dynamic ONNX (

export.batch_size: -1

) both support

fp16

; height and width are always pinned to the trace shape (H/W-dynamic engines are not supported — build separate engines per (H, W)). For the NGC release (576×960), set

export.batch_size: 1

export.opset_version: 17

export.on_cpu: True

See

references/foundation-stereo-export-trt-hardware.md

for the full export / TRT defaults (the opset-vs-

on_cpu

pairing rules, determinism notes,

on_cpu

GPU-memory thresholds) and the Hardware requirements. See

references/tao-deploy-foundation-stereo.md

for the three supported deploy paths and the validation table.

Full TAO Deploy reference: tao-deploy-foundation-stereo.

TRT数据类型为FP32/FP16。静态形状ONNX（

export.batch_size: 1

）和仅批量动态ONNX（

export.batch_size: -1

）均支持

fp16

；高度和宽度始终固定为跟踪形状（不支持H/W动态引擎——需为每个(H, W)构建单独的引擎）。对于NGC版本（576×960），设置

export.batch_size: 1

、

export.opset_version: 17

、

export.on_cpu: True

。

完整的导出/TRT默认值（opset与

on_cpu

的配对规则、确定性说明、

on_cpu

GPU内存阈值）以及硬件要求，请参阅

references/foundation-stereo-export-trt-hardware.md

。三种支持的部署路径和验证表，请参阅

references/tao-deploy-foundation-stereo.md

。

完整TAO Deploy参考：tao-deploy-foundation-stereo。

Error Patterns

错误模式

Common issues: disparity overflow (reduce

model.max_disparity

); missing pretrained paths (set both

model.stereo_backbone.depth_anything_v2_pretrained_path

and

model.stereo_backbone.edgenext_pretrained_path

);

Key 'encoder' not in 'StereoBackBone'

(

encoder

is top-level

model.encoder

);

Key 'dataset_name' is not in struct

(each

data_sources

entry needs both

data_file

and

dataset_name

);

bash: exec: depth_net_stereo: not found

(entrypoint is

depth_net

, no suffix).

See

references/foundation-stereo-troubleshooting.md

for the full error patterns plus the pyt-vs-deploy

crop_size

discussion (the pyt

evaluate

path runs at native image resolution and ignores

crop_size

, with the Middlebury resolution guidance) and the Shape consistency rule.

常见问题：视差溢出（减小

model.max_disparity

）；缺失预训练路径（设置

model.stereo_backbone.depth_anything_v2_pretrained_path

和

model.stereo_backbone.edgenext_pretrained_path

）；

Key 'encoder' not in 'StereoBackBone'

（

encoder

是顶层的

model.encoder

）；

Key 'dataset_name' is not in struct

（每个

data_sources

条目需要

data_file

和

dataset_name

）；

bash: exec: depth_net_stereo: not found

（入口点是

depth_net

，无后缀）。

完整的错误模式，以及pyt与部署端

crop_size

的讨论（pyt的

evaluate

路径以原生图像分辨率运行，忽略

crop_size

，并提供Middlebury分辨率指南）和形状一致性规则，请参阅

references/foundation-stereo-troubleshooting.md

。

Spec Param / Parent Model Inference

规范参数/父模型推理

Model-specific inference mappings belong in MD, not in

config.json

. Generated runners read these mappings and apply them with SDK helpers before

create_job()

(mirrors the old microservices

infer_params.py

flow). For

parent_model

parent_model_folder

, pass the upstream train/export/AutoML child job id as

parent_job_id

; the SDK lists the parent result folder, filters checkpoint artifacts, and returns the selected model file or folder. Do not add these mappings back to

config.json

and do not patch generated runner scripts to guess checkpoint paths.

See

references/foundation-stereo-spec-param-inference.md

for the full per-action inference-mapping table (train / evaluate / inference / export / gen_trt_engine / quantize, including the train pretrained-path link/destination and resume-checkpoint mappings).

模型特定的推理映射应放在MD文件中，而非

config.json

。生成的运行器会读取这些映射，并在

create_job()

前使用SDK助手应用它们（镜像旧微服务的

infer_params.py

流程）。对于

parent_model

parent_model_folder

，将上游训练/导出/AutoML子任务ID作为

parent_job_id

传递；SDK会列出父结果文件夹，过滤检查点产物，并返回选中的模型文件或文件夹。不要将这些映射添加回

config.json

，也不要修改生成的运行器脚本来猜测检查点路径。

完整的各操作推理映射表（train/evaluate/inference/export/gen_trt_engine/quantize，包括训练预训练路径链接/目标和恢复检查点映射），请参阅

references/foundation-stereo-spec-param-inference.md

。

tao-train-foundation-stereo

Original

Translation

Depth Net Stereo

Depth Net Stereo

Train Action Policy

训练操作策略

Workflow

工作流

Prerequisites — data accessibility

前提条件——数据可访问性

Step 1 — Annotation file

步骤1——标注文件

Step 2 — Pair
`model_type`
and
`dataset_name`
based on your data

步骤2——根据数据匹配
`model_type`
和
`dataset_name`

Step 3 — Write spec yaml from Typical Spec Overrides

步骤3——根据典型规范覆盖编写spec yaml

Step 4 — Run

步骤4——运行

Step 5 — Verify

步骤5——验证

Training Requirements

训练要求

Per-Action Dataset Requirements

各操作的数据集要求

Typical Spec Overrides

典型规范覆盖

Eval Dataset

评估数据集

Important Parameters

重要参数

Multi-GPU / Multi-Node

多GPU/多节点

Export / TRT Defaults

导出/TRT默认值

Error Patterns

错误模式

Spec Param / Parent Model Inference

规范参数/父模型推理

tao-train-foundation-stereo

Original

Translation

Depth Net Stereo

Depth Net Stereo

Train Action Policy

训练操作策略

Workflow

工作流

Prerequisites — data accessibility

前提条件——数据可访问性

Step 1 — Annotation file

步骤1——标注文件

Step 2 — Pair model_type and dataset_name based on your data

步骤2——根据数据匹配model_type和dataset_name

Step 3 — Write spec yaml from Typical Spec Overrides

步骤3——根据典型规范覆盖编写spec yaml

Step 4 — Run

步骤4——运行

Step 5 — Verify

步骤5——验证

Training Requirements

训练要求

Per-Action Dataset Requirements

各操作的数据集要求

Typical Spec Overrides

典型规范覆盖

Eval Dataset

评估数据集

Important Parameters

重要参数

Multi-GPU / Multi-Node

多GPU/多节点

Export / TRT Defaults

导出/TRT默认值

Error Patterns

错误模式

Spec Param / Parent Model Inference

规范参数/父模型推理

Step 2 — Pair
`model_type`
and
`dataset_name`
based on your data

步骤2——根据数据匹配
`model_type`
和
`dataset_name`