AlexNet was Fine Tuned to a binary classification task (Chair/Wine bottle)
-
Pretrained AlexNet model was first loaded using torchvision.
-
Uploaded the images I am interested in classifying (Chair/Wine bottle).
-
Randomly Augmented these images to increase sample size and variation (PIL).
-
Created Custom Dataset and DataLoader.
-
All the layers in AlexNet were frozen (requires_grad=False).
-
The last Linear Layer was replaced with a Sequential Layer (Linear and ReLU) with a drop out of .4.
-
This was fed to a sigmoid for my binary classification (Chair/Wine bottle).
-
Trained the newly added Sequential layer and Validated my model.