How to use this approach in case of resnet architecture

kumar-shridhar / PyTorch-BayesianCNN

Bayesian Convolutional Neural Network with Variational Inference based on Bayes by Backprop in PyTorch.

MIT License

1.42k stars 323 forks source link

How to use this approach in case of resnet architecture #51

Open surekhag28 opened 4 years ago

surekhag28 commented 4 years ago

This is not an issue but not sure where should I ask this question hence thought of posting it here. From the models I can see that it is very linear and do not need forward function as it has common wrapper. I tried to follow this approach for resnet architecture but I am kind of end up getting always 14% accuracy. Will it be possible for you to provide sample example for the same (just the model)?

lvyilin commented 4 years ago

I got the same problem while implement in resnet50 and train by imagenet. It got about 23% acc in 40 epochs. It seems that hard to train in deeper architecture.