Embedded Files
Q: Did you use backpropagation to fine-tune some of layers of the CNN? If so, what layers were they? What learning rate did you use? How did you check that it was doing something useful (e.g. loss decreasing across epochs)?
A: In the accuracy graph, we have shown that across epochs, we have had decreasing training and validation loss.
Q: Did your retrained CNN work? Did the results come out as well as you expected?
A: Yes it worked as anticipated . The dataset we used was unbalanced ( class 0 - no airplanes ,was much larger than class 1- airplanes. Due to this, the test accuracy was lower than the accuracy of training and validation.
Q: Are there any lessons learned that would cause you to do something differently next time?
A: We learnt that if we were to use a balanced database, we would achieve better accuracy. Secondly, data-augmentation and normalization helped us avoid overfitting and we were able to process a variety of input images
Google Sites
Report abuse