You could try the following -

1. Reduce the batch size for more epochs.

2. Decrease the learning rate a bit more, it is clearly overfitting.

3. Include Batch normalisation layers or group normalisation.

4. Try a different architecture? Maybe Densenet?

5. Include dropout.

6. Pre-process images. Make sure there is proper distribution in test train and validation.

Thank you so much, but please tell me how to include droput and how to introduce batch normalization.

what kind of images u are trying to train that also depends. ur system is suffering overfitting issue. the number of images for training and validation also matters. Initial learning rate too low in ur case. start trying from 0.001 and decrease to 0.0001. First try with small no. of epochs like 30, 50 etc. I think u tried single-fold training and validation. Increase the no. of fold.

Thank you so much Madam

Thanks you so much

Please tell me how to preprocess and normalize 2d images ?

The strange part is I have tried everything. I have even preprocessed the training images but when i am writing the same code to preprocess the test images, I am being giving the error :-

image = mat2gray(image); ##newly added but this line has worked for the training set.

Even after training for 500 epochs by changing the learning rates weight decay in case of resnet50 the error curve is only given by:-the training accuracy reaches 100% in 100 epochs, but validation accuracy is still at 76%, & no test images are being recognized.

Any help will be appreciated, because I have tried practically almost everything.

My email is:[email protected]

Thanks

P.Bagchi

Drop out and and batch normalisation are in-built functions in TF and PyTorch. You can just import their libraries and call them.

thanks

I have reduced my training classes to 2, but after doing a lot of things like changing learning rates, weight decay, still i am not getting good training and validation accuracies.

What should I do pl. suggest??????????