VGG part of loss function cannot be decreased at all.

Nikronic / CoarseNet

Modified version of U-net based on "Convolutional Networks for Biomedical Image Segmentation (Ronneberger et al., 2015)" paper.

https://arxiv.org/abs/1505.04597

MIT License

25 stars 2 forks source link

Open Nikronic opened 5 years ago

Nikronic commented 5 years ago

The generated output of the model regarding only this loss, is a full noise even after unnormalization.

[ ] Write a function to create a feature map of contributed layers of VGG to the loss using Tensorboad.
[ ] Enrich implemented function in previous step to its job for all other part of the model.
[ ] Use other options of Tensorboad to create a detailed report of issues and progresses.
[ ] Update readme with respect to the obtained report.

Nikronic commented 3 years ago

Maybe computational graph is broken there?