Can't reproduce results in the paper

huyquoctrinh / MetaPolyp-CBMS2023

Meta-Polyp: a baseline for efficient Polyp segmentation (CBMS 2023)

MIT License

68 stars 9 forks source link

Open hungk64it1x opened 4 months ago

hungk64it1x commented 4 months ago

Hello Huy,

After I tried 3 times to train your model (from scratch or with pretrained model) I couldn't reproduce your score (Kvasir score seems to be got highest ~ 0.90x) sometimes It shows val_loss did not improve from epoch 2 (https://drive.google.com/file/d/1Y881vzx65M2MR0z0Gfd-MoPCPuGrMeml/view?fbclid=IwAR1M1-pNwNNdcmrM8N1IrhGhhl5TsDcuHBHmiObcvgeXwARqgBUetvmwXfk)

I provided the weight I have trained.

Thank you very much.

huyquoctrinh commented 4 months ago

Hi, thank you so much, I will reproduce your result again