Could not achieve accuracy of more than 32%

I applied deep learning to 10 different species of monkey. with cnn i reached accuracy of 37% and I tried resnet9 and was able to get to only 32%. what can I do to reach atleast 70%.
Thanks in advance

Few things you could try to improve the accuracy.

  1. Apply different transformations, like normalization, random crop, random horizontal flips, colorjitter etc.
  2. Use a higher value of learning rate like 0.01 or 0.05.
  3. Use bigger architectures of models like resnet18 or resnet34 (pre-trained model).

Since there are not enough images to train a model, I would recommend using data augmentation or try to use a pretrained model (transfer learning).

1 Like