tao-train-ocdnet

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

OCDNet

OCDNet for scene text detection. Detects arbitrary-oriented text regions in natural images using a differentiable binarization approach.

Set train.pretrained_model_path for pretrained weights.

For TAO Deploy TensorRT actions (

gen_trt_engine

, TensorRT

evaluate

, and TensorRT

inference

), read

references/tao-deploy-ocdnet.md

first. Deploy spec templates live in this skill's

references/

folder with the

spec_template_deploy_*.yaml

prefix.

OCDNet用于场景文本检测。它采用可微分二值化方法检测自然图像中的任意方向文本区域。

设置train.pretrained_model_path以加载预训练权重。

对于TAO Deploy TensorRT操作（

gen_trt_engine

、TensorRT

evaluate

和TensorRT

inference

），请先阅读

references/tao-deploy-ocdnet.md

。部署规格模板存放在该技能的

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

列出了可用操作。每个生成的schema还会从schema顶层的

default

字段生成

references/spec_template_<action>.yaml

。AutoML支持在

references/skill_info.yaml

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/最大边界、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

；维护人员会在打包技能库前重新生成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，请先读取

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

；否则默认使用

auto

。当

automl_policy: auto

、

automl_enabled: true

且

schemas/train.schema.json

和

references/spec_template_train.yaml

已打包时，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，直到生成对应的schema。

非训练操作（如

evaluate

、

inference

、

export

以及部署流程）仍在该模型技能中执行。每次运行的

automl_policy

覆盖配置不会更改模型元数据。

Training Requirements

训练要求

Dataset type: ocdnet
Formats: default
Monitoring metric: hmean

数据集类型： ocdnet
格式： default
监控指标： hmean

Per-Action Dataset Requirements

各操作的数据集要求

Action	Spec Key	Source	Files	List?
evaluate	dataset.validate_dataset.data_path	eval_dataset	test.tar.gz	Yes
gen_trt_engine	gen_trt_engine.tensorrt.calibration.cal_image_dir	calibration_dataset	train/img.tar.gz	Yes
inference	inference.input_folder	eval_dataset	test/img.tar.gz	No
prune	dataset.validate_dataset.data_path	eval_dataset	test.tar.gz	Yes
quantize	dataset.train_dataset.data_path	train_datasets	train.tar.gz	Yes
quantize	dataset.validate_dataset.data_path	eval_dataset	test.tar.gz	Yes
quantize	dataset.quant_calibration_dataset.images_dir	train_datasets	train/img.tar.gz	No
retrain	dataset.train_dataset.data_path	train_datasets	train.tar.gz	Yes
retrain	dataset.validate_dataset.data_path	eval_dataset	test.tar.gz	Yes
train	dataset.train_dataset.data_path	train_datasets	train.tar.gz	Yes
train	dataset.validate_dataset.data_path	eval_dataset	test.tar.gz	Yes

操作	规格键	来源	文件	是否为列表？
evaluate	dataset.validate_dataset.data_path	eval_dataset	test.tar.gz	是
gen_trt_engine	gen_trt_engine.tensorrt.calibration.cal_image_dir	calibration_dataset	train/img.tar.gz	是
inference	inference.input_folder	eval_dataset	test/img.tar.gz	否
prune	dataset.validate_dataset.data_path	eval_dataset	test.tar.gz	是
quantize	dataset.train_dataset.data_path	train_datasets	train.tar.gz	是
quantize	dataset.validate_dataset.data_path	eval_dataset	test.tar.gz	是
quantize	dataset.quant_calibration_dataset.images_dir	train_datasets	train/img.tar.gz	否
retrain	dataset.train_dataset.data_path	train_datasets	train.tar.gz	是
retrain	dataset.validate_dataset.data_path	eval_dataset	test.tar.gz	是
train	dataset.train_dataset.data_path	train_datasets	train.tar.gz	是
train	dataset.validate_dataset.data_path	eval_dataset	test.tar.gz	是

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": 30,
    "train.checkpoint_interval": 10,
    "train.validation_interval": 10,
    "train.num_gpus": 1,
    "dataset.train_dataset.loader.batch_size": 16,
    "dataset.train_dataset.data_path": [f"{S3_TRAIN}/train.tar.gz"],
    "dataset.validate_dataset.data_path": [f"{S3_EVAL}/test.tar.gz"],
}

gen_trt_engine (mandatory data sources):

python

{
    "gen_trt_engine.tensorrt.data_type": "INT8",
    "gen_trt_engine.tensorrt.calibration.cal_image_dir": [f"{S3_TRAIN}/train/img.tar.gz"],
}

evaluate (mandatory data sources):

python

{
    "dataset.validate_dataset.data_path": [f"{S3_EVAL}/test.tar.gz"],
}

inference (mandatory data sources):

python

{
    "inference.input_folder": f"{S3_EVAL}/test/img.tar.gz",
}

prune (mandatory data sources):

python

{
    "dataset.validate_dataset.data_path": [f"{S3_EVAL}/test.tar.gz"],
}

quantize (mandatory data sources):

python

{
    "dataset.train_dataset.data_path": [f"{S3_TRAIN}/train.tar.gz"],
    "dataset.validate_dataset.data_path": [f"{S3_EVAL}/test.tar.gz"],
    "dataset.quant_calibration_dataset.images_dir": f"{S3_TRAIN}/train/img.tar.gz",
}

retrain (mandatory data sources):

python

{
    "dataset.train_dataset.data_path": [f"{S3_TRAIN}/train.tar.gz"],
    "dataset.validate_dataset.data_path": [f"{S3_EVAL}/test.tar.gz"],
}

数据源覆盖配置对每个操作都是必填项——Agent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。

python

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

train（必填数据源）：

python

{
    "train.num_epochs": 30,
    "train.checkpoint_interval": 10,
    "train.validation_interval": 10,
    "train.num_gpus": 1,
    "dataset.train_dataset.loader.batch_size": 16,
    "dataset.train_dataset.data_path": [f"{S3_TRAIN}/train.tar.gz"],
    "dataset.validate_dataset.data_path": [f"{S3_EVAL}/test.tar.gz"],
}

gen_trt_engine（必填数据源）：

python

{
    "gen_trt_engine.tensorrt.data_type": "INT8",
    "gen_trt_engine.tensorrt.calibration.cal_image_dir": [f"{S3_TRAIN}/train/img.tar.gz"],
}

evaluate（必填数据源）：

python

{
    "dataset.validate_dataset.data_path": [f"{S3_EVAL}/test.tar.gz"],
}

inference（必填数据源）：

python

{
    "inference.input_folder": f"{S3_EVAL}/test/img.tar.gz",
}

prune（必填数据源）：

python

{
    "dataset.validate_dataset.data_path": [f"{S3_EVAL}/test.tar.gz"],
}

quantize（必填数据源）：

python

{
    "dataset.train_dataset.data_path": [f"{S3_TRAIN}/train.tar.gz"],
    "dataset.validate_dataset.data_path": [f"{S3_EVAL}/test.tar.gz"],
    "dataset.quant_calibration_dataset.images_dir": f"{S3_TRAIN}/train/img.tar.gz",
}

retrain（必填数据源）：

python

{
    "dataset.train_dataset.data_path": [f"{S3_TRAIN}/train.tar.gz"],
    "dataset.validate_dataset.data_path": [f"{S3_EVAL}/test.tar.gz"],
}

Eval Dataset

评估数据集

Optional. Test dataset provided as separate tarball.

可选。测试数据集以单独的tar包形式提供。

Important Parameters

重要参数

model.backbone: Default deformable_resnet18. Deformable convolutions improve text region detection for irregular text.
train.optimizer.args.lr: Learning rate. Default 0.001 (Adam).
postprocess.thresh: Binarization threshold for text region extraction.
postprocess.box_thresh: Box confidence threshold for filtering detections.

model.backbone: 默认值为deformable_resnet18。可变形卷积提升了不规则文本的区域检测效果。
train.optimizer.args.lr: 学习率。默认值为0.001（Adam优化器）。
postprocess.thresh: 文本区域提取的二值化阈值。
postprocess.box_thresh: 过滤检测结果的框置信度阈值。

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.distributed_strategy`	`ddp` , `fsdp` , or `deepspeed_stage_3_offload`	`ddp`

```
ddp
```
with activation checkpointing:
```
find_unused_parameters=False
```
```
ddp
```
without:
```
find_unused_parameters=True
```
```
fsdp
```
forces FP16
deepspeed_stage_3_offload
is uniquely supported for OCDNet (forces FP16)
FAN backbones auto-enable
```
sync_batchnorm
```

启动方式： Lightning管理（单个

python

进程，Lightning生成工作进程）。

规格键	描述	默认值
`train.num_gpus`	GPU数量	1
`train.gpu_ids`	GPU设备索引	[0]
`train.distributed_strategy`	`ddp` 、 `fsdp` 或 `deepspeed_stage_3_offload`	`ddp`

带激活检查点的
```
ddp
```
：
```
find_unused_parameters=False
```
不带激活检查点的
```
ddp
```
：
```
find_unused_parameters=True
```
```
fsdp
```
强制使用FP16
**
```
deepspeed_stage_3_offload
```
**是OCDNet独支持的策略（强制使用FP16）
FAN骨干网络自动启用
```
sync_batchnorm
```

Hardware

硬件要求

Minimum 1 GPU(s), recommended 1 GPU(s). 8GB+ VRAM per GPU. OCDNet is lightweight. Single GPU is sufficient for most datasets.

最低要求1块GPU，推荐使用1块GPU。每块GPU需8GB以上显存。OCDNet模型轻量化，单GPU即可处理大多数数据集。

Error Patterns

错误模式

Low detection rate: Tune postprocess.thresh and box_thresh. Default thresholds may be too aggressive for some datasets.

检测率低：调整postprocess.thresh和box_thresh参数。默认阈值可能对部分数据集过于严格。

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

ocdnet.config.json

Action	Spec Field	Inference Function	Meaning
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	`model.pruned_graph_path`	`pruned_model`	parent pruned model
evaluate	`results_dir`	`output_dir`	current job results directory
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	`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	`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	`model.pruned_graph_path`	`pruned_model`	parent pruned model
inference	`results_dir`	`output_dir`	current job results directory
prune	`prune.checkpoint`	`parent_model`	model file inferred from the parent job results folder
prune	`results_dir`	`output_dir`	current job results directory
quantize	`quantize.model_path`	`parent_model`	model file inferred from the parent job results folder
quantize	`results_dir`	`output_dir`	current job results directory
retrain	`model.pruned_graph_path`	`parent_model`	model file inferred from the parent job results folder
retrain	`results_dir`	`output_dir`	current job results directory
train	`model.pretrained_model_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，并在调用

create_job()

前使用SDK助手应用这些映射。这与旧微服务的

infer_params.py

流程一致。

来自TAO Core

ocdnet.config.json

的推理映射：

操作	规格字段	推理函数	含义
evaluate	`evaluate.checkpoint`	`parent_model`	从父作业结果文件夹推断的模型文件
evaluate	`evaluate.trt_engine`	`parent_model`	从父作业结果文件夹推断的模型文件
evaluate	`model.pruned_graph_path`	`pruned_model`	父剪枝模型
evaluate	`results_dir`	`output_dir`	当前作业结果目录
export	`export.checkpoint`	`parent_model`	从父作业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`	当前作业结果目录
gen_trt_engine	`gen_trt_engine.onnx_file`	`parent_model`	从父作业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`	当前作业结果目录
inference	`inference.checkpoint`	`parent_model`	从父作业结果文件夹推断的模型文件
inference	`inference.trt_engine`	`parent_model`	从父作业结果文件夹推断的模型文件
inference	`model.pruned_graph_path`	`pruned_model`	父剪枝模型
inference	`results_dir`	`output_dir`	当前作业结果目录
prune	`prune.checkpoint`	`parent_model`	从父作业结果文件夹推断的模型文件
prune	`results_dir`	`output_dir`	当前作业结果目录
quantize	`quantize.model_path`	`parent_model`	从父作业结果文件夹推断的模型文件
quantize	`results_dir`	`output_dir`	当前作业结果目录
retrain	`model.pruned_graph_path`	`parent_model`	从父作业结果文件夹推断的模型文件
retrain	`results_dir`	`output_dir`	当前作业结果目录
train	`model.pretrained_model_path`	`ptm_if_no_resume_model`	无恢复检查点时使用预训练模型
train	`results_dir`	`output_dir`	当前作业结果目录
train	`train.resume_training_checkpoint_path`	`resume_model`	从当前作业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子作业ID作为

parent_job_id

传入。SDK会列出父结果文件夹，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。

Deployment

部署

tao-deploy-ocdnet — OCDNet deploy workflow for TensorRT engine generation, TensorRT evaluation, and TensorRT inference using TAO Deploy.

tao-deploy-ocdnet —— 用于TensorRT引擎生成、TensorRT评估和TensorRT推理的OCDNet部署工作流，基于TAO Deploy实现。