different result between batch evaluate and per image predict

[yinchenghao]

when I evaluate the batch images and test the same image, the result is different, evaluate accuracy is almost 0.98,but the same image test result is bad,what's wrong?Here is my evaluate result and test result,my training dataset is chinese

accuracy and evaluate results:

test result:

[YooWang]

I am also studying this project. Can we communicate with each other? In wechat

yinchenghao notifications@github.com于2020年7月17日 周五12:53写道：

when I evaluate the batch images and test the same image,but the result is different,the evaluate accuracy is almost 0.98,but the same image test result is bad,what's wrong?Here is my evaluate result and test result,my training dataset is chinese

accuracy and evaluate results: [image: accuracy] https://user-images.githubusercontent.com/38687338/87746776-09fa5500-c824-11ea-8ca8-4105f5c8b561.png

[image: 1] https://user-images.githubusercontent.com/38687338/87746738-e9ca9600-c823-11ea-974d-54b0e5fb4d17.png

test result: [image: 2] https://user-images.githubusercontent.com/38687338/87746795-1088cc80-c824-11ea-982c-98964c4d73f8.png [image: 3] https://user-images.githubusercontent.com/38687338/87746800-1383bd00-c824-11ea-811c-9aaf516a634c.png [image: 4] https://user-images.githubusercontent.com/38687338/87746804-154d8080-c824-11ea-82a9-c29d141ba5ff.png

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/ayumiymk/aster.pytorch/issues/51, or unsubscribe https://github.com/notifications/unsubscribe-auth/AOF47RLDLBT6NI3LARKR75DR37DLPANCNFSM4O5OKOMQ .

[yinchenghao]

I am also studying this project. Can we communicate with each other? In wechat yinchenghao notifications@github.com于2020年7月17日 周五12:53写道： … when I evaluate the batch images and test the same image,but the result is different,the evaluate accuracy is almost 0.98,but the same image test result is bad,what's wrong?Here is my evaluate result and test result,my training dataset is chinese accuracy and evaluate results: [image: accuracy] https://user-images.githubusercontent.com/38687338/87746776-09fa5500-c824-11ea-8ca8-4105f5c8b561.png [image: 1] https://user-images.githubusercontent.com/38687338/87746738-e9ca9600-c823-11ea-974d-54b0e5fb4d17.png test result: [image: 2] https://user-images.githubusercontent.com/38687338/87746795-1088cc80-c824-11ea-982c-98964c4d73f8.png [image: 3] https://user-images.githubusercontent.com/38687338/87746800-1383bd00-c824-11ea-811c-9aaf516a634c.png [image: 4] https://user-images.githubusercontent.com/38687338/87746804-154d8080-c824-11ea-82a9-c29d141ba5ff.png — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#51>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AOF47RLDLBT6NI3LARKR75DR37DLPANCNFSM4O5OKOMQ .

of course， my wechat:YCH100289

[ayumiymk]

I am not clear about what happened. Good luck to both of you!