Lucas Vazquez
Lucas Vazquez
[This](https://arxiv.org/pdf/1706.03319.pdf) paper does a great explanation of why U-net might fail in some cases. Quoting from the paper: _The U-net is ”lazy”. That is to say if the U-net find...
A strategy for solving the issue can be: * Freeze skip connections and train network * After *some time* unfreeze skip connections and see what happens?
The paper talks about "Guide decoders", although it's not deeply explained what they mean. I think what I can try doing, is generating the image without the skip connections at...
First try to modify `DynamicUnet` failed miserably. Need to find a away to get the output of each `UnetBlock` with and without skip connections, and then use that to create...
Can also select a random box, scale it, and place somewhere else in the image
Can use `ShortEpochCallback`
This issue is actually getting more important. `FeatModels` works well for style transfer, but it actually doesn't makes sense for other tasks e.g. superres
Actually, there is no current easy way of doing this, can you describe the solution you would like? Maybe a function that take the model, a dataloader, and the metrics?
@FraPochetti is this still a thing?
Hey Adam, this issue makes a very good point on why we should not use dicts as metrics for fastai, I'm on board with that. The current implementation (using dicts)...