ReaFly / ACSNet

MICCAI 2020 : Adaptive Context Selection for Polyp Segmentation (Pytorch implementation).
52 stars 4 forks source link

How to reproduce the reported results? #7

Open zhou07 opened 2 years ago

zhou07 commented 2 years ago

Hi, thanks for releasing the code. I appreciate the proposed ACSNet so much and want to take it as my baseline to conduct some further explorations. Following the instructions, I run the code for Kvasir_SEG. However, I failed to reproduce the reported results. mIoU: reproduce: 90.59<--> report: 91.40 ACC: reproduce: 97.13<--> report: 97.64 Dice: reproduce: 90.04<--> report: 91.30

And above reproduced results are selected by testing all the 150 checkpoints and choosing the best one. I have run the code several times, but the results are still far from the reported ones. Would you mind telling me how to reproduce the result? And How do you choose the proper checkpoint for testing?