Project 2

Image Classification with CNN


1. Direct prediction

  • Download the dataset birds from http://www-cvr.ai.uiuc.edu/ponce_grp/data/.
  • Use Keras and a CNN from Keras Applications pretrained on ImageNet, to classify the images in the birds dataset. Construct a confusion matrix that relates the bird classes with the 10 most frequent classes from ImageNet predicted by the model.
  • Discuss the results.

2. Transfer learning

  • Use the pre-trained CNN model as a feature extractor. Create a new model that replaces the top part of the pretrained CNN with two layers of 256 and 6 neurons respectively.
  • Train the model with the training images from the bird dataset.
  • Evaluate the performance over the test dataset reporting the results in a confusion matrix. Discuss the results.

3. Fine tuning

  • Repeat the experiment from the last question, but this time allow all the layers to be trained.
  • Compare and discuss the results.