ixrt

DBNet (ixRT)

Model Description

DBNet (Differentiable Binarization Network) is a scene text detection model that uses a differentiable binarization process for robust text detection.

Supported Environments

GPU	IXUCA SDK	Release	Branch
MR-V100	4.4.0	26.06	release/26.06

Note: 请切换到与您的 SDK 版本对应的 Release 分支进行测试。请勿直接在 master 分支上运行测试，因为 master 分支可能包含与您的本地 SDK 版本不兼容的最新更改。

切换分支命令示例：git checkout release/26.06

Model Preparation

Prepare Resources

Pretrained models:

• r50_en_dbnet: http://files.deepspark.org.cn:880/deepspark/data/checkpoints/r50_en_dbnet.onnx

Dataset: ICDAR 2015 http://files.deepspark.org.cn:880/deepspark/data/datasets/icdar_2015.zip

Install Dependencies

pip3 install shapely pyclipper opencv-python==4.6.0.66 tqdm

Model Conversion

mkdir checkpoints
cd checkpoints
wget http://files.deepspark.org.cn:880/deepspark/data/checkpoints/r50_en_dbnet.onnx

Model Inference

export DATASETS_DIR=/path/to/icdar2015/
export CHECKPOINTS_DIR=./checkpoints
export RUN_DIR=./

FP16

# Test ACC
bash scripts/infer_dbnet_fp16_accuracy.sh
# Test FPS
bash scripts/infer_dbnet_fp16_performance.sh

INT8

# Test ACC
bash scripts/infer_dbnet_int8_accuracy.sh
# Test FPS
bash scripts/infer_dbnet_int8_performance.sh

Model Results

Model	Backbone	BatchSize	Precision	FPS	Hmean
DBNet	r50_en	32	FP16	143.85	0.803
DBNet	r50_en	32	INT8	143.73	0.803