Develop computer vision model to classify images of ant and bees using transfer learning.
Training large networks from scratch with limited resources and/or limited data sets is not pragmatic. Recognizing this, Transfer Learning is quite popular for vision applications with pre-trained encoders and language applications with pre-trained language models.
Here use two major transfer learning scenarios for our vision model:
- Finetuning the ConvNet Instead of random initializaion, we initialize the network with a pretrained network, like the one that is trained on imagenet 1000 dataset. Rest of the training looks as usual.
- ConvNet as fixed feature extractor We freeze the weights for all of the network except that of the final fully connected layer. The last fully connected layer is replaced with a new one with random weights and only this layer is trained.
We will use torchvision and torch.utils.data packages for loading the data.
We are training a model to classify ants and bees. We have about 120 training images each for ants and bees. There are 75 validation images for each class. Usually, this is a very small dataset to generalize upon, if trained from scratch. Since we are using transfer learning, we should be able to generalize reasonably well.
This dataset is a very small subset of imagenet. Download and extract it to the current directory. https://download.pytorch.org/tutorial/hymenoptera_data.zip
- Augment and normalize the training data
- Normalize the validation data