/East.Pytorch

Primary LanguagePythonMIT LicenseMIT

East.Pytorch

This is a PyTorch Re-Implementation of EAST

Performance on ICDAR2015

download pretrain model east_82,password is 54ck,this model was pretraintrained by VISD 10K images,then finetuned by ICDAR15 1K images

model precision recall F1
paper(VGG16+RBOX) 0.8046 0.7275 0.7641
Our(resnet18+RBOX) 0.835 0.7554 0.7932
Our(VISD10K+IC15 1K) 0.883 0.767 0.821

How to train this model

1.Modify image path and gt path in train.py

2.run train.py

How to test images

1.Modify image path and model path in detect.py

2.run detect.py,the results are saved in "res" folder

How to eval result

1.Modify image path and model path in eval.py

2.run deval.py and you will get the result

keep update