unet

[wcx-bit]

dear author, the unet model you implemented seems different with that in the original paper. The initial features is 64 which is so big. meanwhile, the original unet should be input channel to 32 then to 64 in the first layer. while you just skip the middle and to the 64 at first, then you use a 3x3 conv without change channel, I think there is some difference with others original model

[Yanfeng-Zhou]

dear author, the unet model you implemented seems different with that in the original paper. The initial features is 64 which is so big. meanwhile, the original unet should be input channel to 32 then to 64 in the first layer. while you just skip the middle and to the 64 at first, then you use a 3x3 conv without change channel, I think there is some difference with others original model

Yes, my implementation of UNet is a little different from the original paper. You can adjust the architecture I implemented according to your dataset. The reason I adjusted the UNet architecture is that I hope it will perform better.