SMAI HW Report: Americal Sign Language Identification
Tathagata Raha, 2018114017
From the notebook, we can see that the best accuracy achieved by the AlexNet model on the validation dataset is 0.819 at the end of 14th epoch and the best validation loss is acieved at the end of 17th epoch.
While the accuracy for the model using MLP architecture with Dropout of 30% was 0.333 on training data and 0.261 on test datawith the number of iterations upto 200.
So, we can say the transfer learning from the Alexnet model worked a lot better than training the image classification model from scratch.