The changed evaluation code for comparison with EMANet
The problem encountered is, the predicted mask does not match the loaded label in size some times. The solution we adopted is to unify their sizes using F.interpolate(). Please train the model with dataset and strategies provided in https://github.com/XiaLiPKU/EMANet, and evaluate the trained model with the eval.py provided above.
train_epo11.txt
test_on_Dust_rainy.log
test_on_Night_rainy.log
test_on_Night_sunny.log