weijiawu / scene-text-detection-and-recognition-benchmark

Scene-Text-Detection-and-Recognition-benchmark

Introduction

Scene Text Detection and Recognition is an open source scene text detection and recognition benchmark based on PyTorch

Major features

• Various backbones and pretrained models
• Large-scale training configs
• High efficiency and extensibility

Requirements

This is my experiment eviroument

• python3.6
• pytorch1.6.0+cu101

Todo list

• implement DB
• implement CRNN
• implement ABCNet

Dataset

Detection

Supported:

• ICDAR15
• ICDAR17MLT
• sythtext800K
• Total Text
• CTW1500
• 2019ArT

Recognition

model zoo

Supported text detection:

• EAST EAST: An Efficient and Accurate Scene Text Detector
• Psenet Shape Robust Text Detection with Progressive Scale Expansion Network
• DB Real-time Scene Text Detection with Differentiable Binarization

Supported text recognition:

• CRNN An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application to Scene Text Recognition

Supported End to End:

Installation

Please refer to install.md for installation and dataset preparation.

Experiments

All models are trained in the same condition, and might not get the best result

The performance of EAST

Method	Backbone	Pretrain	Resolution	Dataset	Precision	Recall	F-score	FPS
EAST	VGG16	-	512	ICDAR15	0.81	0.81	0.81	-
EAST	VGG16	-	512	ICDAR17	0.72	0.61	0.66	-
EAST	VGG16	SynthText	512	ICDAR15	0.82	0.824	0.822	-

The performance of PSENet

Method	Backbone	Pretrain	Resolution	Dataset	Precision	Recall	F-score	FPS
PSENet(1s)	ResNet50	-	640*640	SynthText	-	-	-	-
PSENet(1s)	ResNet50	-	640*640	ICDAR15	0.816	0.795	0.805	-
PSENet(1s)	ResNet50	-	640*640	ICDAR17(val)	0.755	0.614	0.677	-
PSENet(1s)	ResNet50	-	640*640	ICDAR17(test)	0.762	0.643	0.698	-
PSENet(1s)	ResNet50	-	640*640	Total-Text	0.8255	0.7597	0.7913	3.0
PSENet(1s)	ResNet50	SynthText	640*640	ICDAR15	0.864	0.835	0.850	-
PSENet(1s)	ResNet50	SynthText+ICDAR17	640*640	ICDAR15	0.883	0.853	0.868	3.0
PSENet(1s)	ResNet50	SynthText	640*640	Total-Text	0.834	0.781	0.807	3.0

links

EAST: https://github.com/SakuraRiven/EAST

PSENet: https://github.com/WenmuZhou/PSENet.pytorch

DB: https://github.com/WenmuZhou/DBNet.pytorch

https://github.com/WenmuZhou/PytorchOCR

Contact

Eamil: [email protected]