search

yingkaisha / keras-unet-collection

The Tensorflow, Keras implementation of U-net, V-net, U-net++, UNET 3+, Attention U-net, R2U-net, ResUnet-a, U^2-Net, TransUNET, and Swin-UNET with optional ImageNet-trained backbones.
MIT License
662 stars 184 forks source link

UNET 3plus not learning on large datasets #41

Open jrmathew1 opened 2 years ago

jrmathew1 commented 2 years ago

I am trying to use the UNET 3plus with deep supervision and cgm for nodule segmentation from lung ct scans. When I train the model on a 1000 image subset of the dataset the model learns and outputs meaningful segmentation masks.

However, when I train on a larger dataset of 2000 or more images the model does not learn and outputs a mask of all 0s always.

I have tried increasing and reducing the learning rate as well changing optimizers and loss functions but the issue remains. I also tried the att_unet_2d model and had the same issue.

I have attached my code: training_code.zip

zhaxideles commented 11 months ago

I also have the same problem, verifying that the loss is a constant, have you solved it？