tao-train-deformable-detr

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Deformable DETR

Deformable DETR for 2D object detection. Uses deformable attention for efficient multi-scale feature processing. Lighter than DINO with competitive accuracy.

Uses pretrained backbone weights. Set model.pretrained_backbone_path for backbone-only loading.

For TAO Deploy TensorRT actions (

gen_trt_engine

, TensorRT

evaluate

, and TensorRT

inference

), read

references/tao-deploy-deformable-detr.md

first. Deploy spec templates live in this skill's

references/

folder with the

spec_template_deploy_*.yaml

prefix.

用于2D目标检测的Deformable DETR。采用可变形注意力实现高效的多尺度特征处理，比DINO更轻量化且精度具有竞争力。

使用预训练骨干网络权重。设置model.pretrained_backbone_path可仅加载骨干网络。

对于TAO Deploy TensorRT操作（

gen_trt_engine

、TensorRT

evaluate

和TensorRT

inference

），请先阅读

references/tao-deploy-deformable-detr.md

。部署规格模板存放在本skill的

references/

文件夹中，前缀为

spec_template_deploy_*.yaml

。

Dataclass Schemas

数据类Schema

Generated TAO Core schemas are packaged in

schemas/<action>.schema.json

, with

schemas/manifest.json

listing available actions. Each generated schema also emits

references/spec_template_<action>.yaml

from the schema top-level

default

field. AutoML enablement is declared at the model layer in

references/skill_info.yaml

via

automl_enabled

. Runnable AutoML still requires

schemas/train.schema.json

and

references/spec_template_train.yaml

to exist and parse. Use the packaged train schema for

automl_default_parameters

automl_disabled_parameters

, defaults, min/max bounds, enums, option weights, math conditions, dependencies, and popular parameters. Do not expect

~/tao-core

at runtime; maintainers regenerate schemas/templates before packaging the skill bank.

生成的TAO Core schema打包在

schemas/<action>.schema.json

中，

schemas/manifest.json

列出了可用的操作。每个生成的schema还会从schema顶层

default

字段生成

references/spec_template_<action>.yaml

。AutoML支持在

references/skill_info.yaml

中的模型层通过

automl_enabled

声明。可运行的AutoML仍要求

schemas/train.schema.json

和

references/spec_template_train.yaml

存在且可解析。使用打包的训练schema来配置

automl_default_parameters

、

automl_disabled_parameters

、默认值、最小/最大范围、枚举值、选项权重、数学条件、依赖关系以及常用参数。运行时不要依赖

~/tao-core

；维护人员在打包skill库前会重新生成schema和模板。

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。处理任何训练阶段请求前，请先阅读

references/skill_info.yaml

，并通过显式的

automl_policy

值或用户的工作流请求确定运行覆盖配置。将“turn off AutoML”、“disable AutoML”、“no HPO”或“plain training”这类短语视为本次运行的

automl_policy: off

；否则默认设为

auto

。当

automl_policy: auto

、

automl_enabled: true

且

schemas/train.schema.json

和

references/spec_template_train.yaml

均已打包时，默认将训练操作通过

tao-skill-bank:tao-run-automl

路由到该模型的

skill_dir

。保留数据集、规格、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schema/模板缺失时，才使用直接模型训练；若schema缺失，需报告该模型已启用AutoML但无法运行，直到生成schema为止。

非训练操作（如

evaluate

、

inference

、

export

和部署流程）仍在本模型skill中执行。每次运行的

automl_policy

覆盖配置不会更改模型元数据。

Training Requirements

训练要求

Dataset type: object_detection
Formats: coco, coco_raw
Monitoring metric: val_mAP50

数据集类型： object_detection
格式： coco, coco_raw
监控指标： val_mAP50

Per-Action Dataset Requirements

各操作的数据集要求

Action	Spec Key	Source	Files	List?
evaluate	dataset.test_data_sources.image_dir	eval_dataset	images.tar.gz	No
evaluate	dataset.test_data_sources.json_file	eval_dataset	annotations.json	No
export	dataset.train_data_sources	train_datasets	image_dir: images.tar.gz, json_file: annotations.json	Yes
export	dataset.val_data_sources	train_datasets	image_dir: images.tar.gz, json_file: annotations.json	Yes
gen_trt_engine	gen_trt_engine.tensorrt.calibration.cal_image_dir	calibration_dataset	images.tar.gz	Yes
inference	dataset.infer_data_sources.image_dir	inference_dataset	images.tar.gz	Yes
inference	dataset.infer_data_sources.classmap	inference_dataset	label_map.txt	No
quantize	dataset.train_data_sources	train_datasets	image_dir: images.tar.gz, json_file: annotations.json	Yes
quantize	dataset.val_data_sources	train_datasets	image_dir: images.tar.gz, json_file: annotations.json	Yes
quantize	dataset.quant_calibration_data_sources	train_datasets	image_dir: images.tar.gz, json_file: annotations.json	No
train	dataset.train_data_sources	train_datasets	image_dir: images.tar.gz, json_file: annotations.json	Yes
train	dataset.val_data_sources	train_datasets	image_dir: images.tar.gz, json_file: annotations.json	Yes

Action	Spec Key	Source	Files	List?
evaluate	dataset.test_data_sources.image_dir	eval_dataset	images.tar.gz	No
evaluate	dataset.test_data_sources.json_file	eval_dataset	annotations.json	No
export	dataset.train_data_sources	train_datasets	image_dir: images.tar.gz, json_file: annotations.json	Yes
export	dataset.val_data_sources	train_datasets	image_dir: images.tar.gz, json_file: annotations.json	Yes
gen_trt_engine	gen_trt_engine.tensorrt.calibration.cal_image_dir	calibration_dataset	images.tar.gz	Yes
inference	dataset.infer_data_sources.image_dir	inference_dataset	images.tar.gz	Yes
inference	dataset.infer_data_sources.classmap	inference_dataset	label_map.txt	No
quantize	dataset.train_data_sources	train_datasets	image_dir: images.tar.gz, json_file: annotations.json	Yes
quantize	dataset.val_data_sources	train_datasets	image_dir: images.tar.gz, json_file: annotations.json	Yes
quantize	dataset.quant_calibration_data_sources	train_datasets	image_dir: images.tar.gz, json_file: annotations.json	No
train	dataset.train_data_sources	train_datasets	image_dir: images.tar.gz, json_file: annotations.json	Yes
train	dataset.val_data_sources	train_datasets	image_dir: images.tar.gz, json_file: annotations.json	Yes

Typical Spec Overrides

典型规格覆盖配置

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

python

S3_TRAIN = "s3://bucket/data/train"
S3_EVAL = "s3://bucket/data/eval"

train (mandatory data sources):

python

{
    "train.num_epochs": 10,
    "train.checkpoint_interval": 10,
    "train.validation_interval": 10,
    "train.num_gpus": 1,
    "dataset.num_classes": "<num_classes> + 1",
    "dataset.train_data_sources": [{"image_dir": f"{S3_TRAIN}/images.tar.gz", "json_file": f"{S3_TRAIN}/annotations.json"}],
    "dataset.val_data_sources": [{"image_dir": f"{S3_TRAIN}/images.tar.gz", "json_file": f"{S3_TRAIN}/annotations.json"}],
}

evaluate (mandatory data sources):

python

{
    "dataset.num_classes": "<num_classes> + 1",
    "dataset.test_data_sources.image_dir": f"{S3_EVAL}/images.tar.gz",
    "dataset.test_data_sources.json_file": f"{S3_EVAL}/annotations.json",
}

export (mandatory data sources):

python

{
    "dataset.num_classes": "<num_classes> + 1",
    "dataset.train_data_sources": [{"image_dir": f"{S3_TRAIN}/images.tar.gz", "json_file": f"{S3_TRAIN}/annotations.json"}],
    "dataset.val_data_sources": [{"image_dir": f"{S3_TRAIN}/images.tar.gz", "json_file": f"{S3_TRAIN}/annotations.json"}],
}

gen_trt_engine (mandatory data sources):

python

{
    "gen_trt_engine.tensorrt.data_type": "FP16",
    "dataset.num_classes": "<num_classes> + 1",
    "gen_trt_engine.tensorrt.calibration.cal_image_dir": [f"{S3_TRAIN}/images.tar.gz"],
}

inference (mandatory data sources):

python

{
    "dataset.num_classes": "<num_classes> + 1",
    "dataset.infer_data_sources.image_dir": [f"{S3_EVAL}/images.tar.gz"],
    "dataset.infer_data_sources.classmap": f"{S3_EVAL}/label_map.txt",
}

quantize (mandatory data sources):

python

{
    "dataset.train_data_sources": [{"image_dir": f"{S3_TRAIN}/images.tar.gz", "json_file": f"{S3_TRAIN}/annotations.json"}],
    "dataset.val_data_sources": [{"image_dir": f"{S3_TRAIN}/images.tar.gz", "json_file": f"{S3_TRAIN}/annotations.json"}],
    "dataset.quant_calibration_data_sources": {"image_dir": f"{S3_TRAIN}/images.tar.gz", "json_file": f"{S3_TRAIN}/annotations.json"},
}

数据源覆盖配置对每个操作都是必需的——Agent必须根据上面的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。

python

S3_TRAIN = "s3://bucket/data/train"
S3_EVAL = "s3://bucket/data/eval"

train（必填数据源）：

python

{
    "train.num_epochs": 10,
    "train.checkpoint_interval": 10,
    "train.validation_interval": 10,
    "train.num_gpus": 1,
    "dataset.num_classes": "<num_classes> + 1",
    "dataset.train_data_sources": [{"image_dir": f"{S3_TRAIN}/images.tar.gz", "json_file": f"{S3_TRAIN}/annotations.json"}],
    "dataset.val_data_sources": [{"image_dir": f"{S3_TRAIN}/images.tar.gz", "json_file": f"{S3_TRAIN}/annotations.json"}],
}

evaluate（必填数据源）：

python

{
    "dataset.num_classes": "<num_classes> + 1",
    "dataset.test_data_sources.image_dir": f"{S3_EVAL}/images.tar.gz",
    "dataset.test_data_sources.json_file": f"{S3_EVAL}/annotations.json",
}

export（必填数据源）：

python

{
    "dataset.num_classes": "<num_classes> + 1",
    "dataset.train_data_sources": [{"image_dir": f"{S3_TRAIN}/images.tar.gz", "json_file": f"{S3_TRAIN}/annotations.json"}],
    "dataset.val_data_sources": [{"image_dir": f"{S3_TRAIN}/images.tar.gz", "json_file": f"{S3_TRAIN}/annotations.json"}],
}

gen_trt_engine（必填数据源）：

python

{
    "gen_trt_engine.tensorrt.data_type": "FP16",
    "dataset.num_classes": "<num_classes> + 1",
    "gen_trt_engine.tensorrt.calibration.cal_image_dir": [f"{S3_TRAIN}/images.tar.gz"],
}

inference（必填数据源）：

python

{
    "dataset.num_classes": "<num_classes> + 1",
    "dataset.infer_data_sources.image_dir": [f"{S3_EVAL}/images.tar.gz"],
    "dataset.infer_data_sources.classmap": f"{S3_EVAL}/label_map.txt",
}

quantize（必填数据源）：

python

{
    "dataset.train_data_sources": [{"image_dir": f"{S3_TRAIN}/images.tar.gz", "json_file": f"{S3_TRAIN}/annotations.json"}],
    "dataset.val_data_sources": [{"image_dir": f"{S3_TRAIN}/images.tar.gz", "json_file": f"{S3_TRAIN}/annotations.json"}],
    "dataset.quant_calibration_data_sources": {"image_dir": f"{S3_TRAIN}/images.tar.gz", "json_file": f"{S3_TRAIN}/annotations.json"},
}

Eval Dataset

评估数据集

Optional. If provided, validation mAP is computed at each checkpoint interval.

可选。若提供，会在每个检查点间隔计算验证mAP。

Important Parameters

重要参数

dataset.num_classes: Number of object classes. Default 91 (COCO). Must match annotations.
model.backbone: Default resnet_50. Supported: resnet_50, gcvit_tiny, gcvit_small, gcvit_base, gcvit_large, gcvit_large_384 (more limited than DINO).
train.optim.lr: Learning rate. Default 2e-4 (AdamW). lr_backbone is 2e-5.
train.optim.lr_steps: MultiStep LR schedule. Default [40]. For short runs, set to match ~80% of total epochs.
model.num_queries: Number of object queries. Default 300. Valid range 100-900.
model.dropout_ratio: Dropout in transformer layers. Default 0.3 (higher than DINO's 0.0). Reduce for large datasets, increase for small datasets.
model.dim_feedforward: FFN hidden dim. Default 1024 (vs DINO's 2048). Increasing improves capacity but costs memory.

dataset.num_classes：目标类别数量。默认91（COCO数据集）。必须与标注匹配。
model.backbone：默认resnet_50。支持的骨干网络：resnet_50, gcvit_tiny, gcvit_small, gcvit_base, gcvit_large, gcvit_large_384（比DINO支持的更少）。
train.optim.lr：学习率。默认2e-4（AdamW优化器）。骨干网络学习率为2e-5。
train.optim.lr_steps：多步学习率调度。默认[40]。对于短训练周期，设置为总周期的~80%。
model.num_queries：目标查询数量。默认300。有效范围100-900。
model.dropout_ratio：Transformer层的 dropout 比例。默认0.3（高于DINO的0.0）。大数据集可降低该值，小数据集可提高该值。
model.dim_feedforward：前馈网络隐藏层维度。默认1024（对比DINO的2048）。增大该值可提升模型能力，但会增加内存消耗。

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 DINO. Multi-node requires

WORLD_SIZE

NODE_RANK

MASTER_ADDR

MASTER_PORT

env vars set by orchestrator.

启动方式： Lightning管理（单个

python

进程，Lightning生成工作进程）。

Spec Key	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`	分布式策略，可选 `ddp` 或 `fsdp`	`ddp`

与DINO的DDP/FSDP行为一致。多节点需要编排器设置

WORLD_SIZE

、

NODE_RANK

、

MASTER_ADDR

、

MASTER_PORT

环境变量。

Export / TRT Defaults

导出/TRT默认值

Export input: 640x640, opset 17
TRT data types: FP32, FP16, INT8
TRT workspace: 1024 MB
TRT max_batch_size: 1

Full TAO Deploy reference: tao-deploy-deformable-detr.

导出输入尺寸：640x640，opset 17
TRT数据类型：FP32, FP16, INT8
TRT工作空间：1024 MB
TRT最大批量大小：1

完整TAO Deploy参考：tao-deploy-deformable-detr。

Hardware

硬件要求

Minimum 1 GPU(s), recommended 4 GPU(s). 16GB+ (V100 or A100) VRAM per GPU. Slightly lighter than DINO due to smaller FFN. batch_size=4 fits on most 16GB+ GPUs.

最低1块GPU，推荐4块GPU。每块GPU需16GB+显存（V100或A100）。由于前馈网络更小，比DINO略轻量化。batch_size=4可适配大多数16GB+显存的GPU。

Error Patterns

错误模式

CUDA out of memory: Reduce batch_size (4 -> 2 -> 1).

num_select must be < num_queries * num_classes: Same constraint as DINO.

return_interm_indices length must match num_feature_levels: Default [1,2,3,4] with num_feature_levels=4.

Dataset size smaller than total batch size: Reduce batch_size or num_gpus.

CUDA内存不足：减小batch_size（4→2→1）。

num_select必须小于num_queries * num_classes：与DINO的约束相同。

return_interm_indices长度必须匹配num_feature_levels：默认[1,2,3,4]，对应num_feature_levels=4。

数据集大小小于总批量大小：减小batch_size或GPU数量。

Spec Param / Parent Model Inference

规格参数/父模型推理

Model-specific inference mappings belong in this MD file, not in

config.json

. Generated runners should read this section and apply the mappings with SDK helpers before

create_job()

. This mirrors the old microservices

infer_params.py

flow.

Inference mappings from TAO Core

deformable_detr.config.json

Action	Spec Field	Inference Function	Meaning
evaluate	`encryption_key`	`key`	encryption key
evaluate	`evaluate.checkpoint`	`parent_model`	model file inferred from the parent job results folder
evaluate	`evaluate.trt_engine`	`parent_model`	model file inferred from the parent job results folder
evaluate	`results_dir`	`output_dir`	current job results directory
export	`encryption_key`	`key`	encryption key
export	`export.checkpoint`	`parent_model`	model file inferred from the parent job results folder
export	`export.onnx_file`	`create_onnx_file`	output ONNX path
export	`results_dir`	`output_dir`	current job results directory
gen_trt_engine	`encryption_key`	`key`	encryption key
gen_trt_engine	`gen_trt_engine.onnx_file`	`parent_model`	model file inferred from the parent job results folder
gen_trt_engine	`gen_trt_engine.tensorrt.calibration.cal_cache_file`	`create_cal_cache`	calibration cache path
gen_trt_engine	`gen_trt_engine.trt_engine`	`create_engine_file`	output TensorRT engine path
gen_trt_engine	`results_dir`	`output_dir`	current job results directory
inference	`encryption_key`	`key`	encryption key
inference	`inference.checkpoint`	`parent_model`	model file inferred from the parent job results folder
inference	`inference.trt_engine`	`parent_model`	model file inferred from the parent job results folder
inference	`results_dir`	`output_dir`	current job results directory
quantize	`encryption_key`	`key`	encryption key
quantize	`quantize.model_path`	`parent_model`	model file inferred from the parent job results folder
quantize	`results_dir`	`output_dir`	current job results directory
train	`encryption_key`	`key`	encryption key
train	`model.pretrained_backbone_path`	`ptm_if_no_resume_model`	PTM when no resume checkpoint exists
train	`results_dir`	`output_dir`	current job results directory
train	`train.resume_training_checkpoint_path`	`resume_model`	model file inferred from the current job results folder

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.

模型特定的推理映射应放在此MD文件中，而非

config.json

。生成的运行器应读取本节内容，并在

create_job()

前使用SDK助手应用映射。这与旧微服务的

infer_params.py

流程一致。

来自TAO Core

deformable_detr.config.json

的推理映射：

Action	Spec Field	Inference Function	含义
evaluate	`encryption_key`	`key`	加密密钥
evaluate	`evaluate.checkpoint`	`parent_model`	从父任务结果文件夹推断的模型文件
evaluate	`evaluate.trt_engine`	`parent_model`	从父任务结果文件夹推断的模型文件
evaluate	`results_dir`	`output_dir`	当前任务结果目录
export	`encryption_key`	`key`	加密密钥
export	`export.checkpoint`	`parent_model`	从父任务结果文件夹推断的模型文件
export	`export.onnx_file`	`create_onnx_file`	输出ONNX路径
export	`results_dir`	`output_dir`	当前任务结果目录
gen_trt_engine	`encryption_key`	`key`	加密密钥
gen_trt_engine	`gen_trt_engine.onnx_file`	`parent_model`	从父任务结果文件夹推断的模型文件
gen_trt_engine	`gen_trt_engine.tensorrt.calibration.cal_cache_file`	`create_cal_cache`	校准缓存路径
gen_trt_engine	`gen_trt_engine.trt_engine`	`create_engine_file`	输出TensorRT引擎路径
gen_trt_engine	`results_dir`	`output_dir`	当前任务结果目录
inference	`encryption_key`	`key`	加密密钥
inference	`inference.checkpoint`	`parent_model`	从父任务结果文件夹推断的模型文件
inference	`inference.trt_engine`	`parent_model`	从父任务结果文件夹推断的模型文件
inference	`results_dir`	`output_dir`	当前任务结果目录
quantize	`encryption_key`	`key`	加密密钥
quantize	`quantize.model_path`	`parent_model`	从父任务结果文件夹推断的模型文件
quantize	`results_dir`	`output_dir`	当前任务结果目录
train	`encryption_key`	`key`	加密密钥
train	`model.pretrained_backbone_path`	`ptm_if_no_resume_model`	无恢复检查点时的预训练模型
train	`results_dir`	`output_dir`	当前任务结果目录
train	`train.resume_training_checkpoint_path`	`resume_model`	从当前任务结果文件夹推断的模型文件

对于

parent_model

或

parent_model_folder

，将上游训练/导出/AutoML子任务ID作为

parent_job_id

传入。SDK会列出父任务结果文件夹，过滤检查点工件，并返回选定的模型文件或文件夹。请勿将这些映射添加回

config.json

，也不要修改生成的运行器脚本以猜测检查点路径。