tuhinaprasad28 / image-feature-extraction-for-classification Goto Github PK

Download the CIFAR 10 dataset (original data can be found here, and here is a link to the pickled python version.

Use the pretrained Resnet18 model (from trochvision) to extract features. Use the features as inputs in a new multi-class logistic regression model (use nn.Linear/ nn.Module to define your model) -(a) Describe any choices made and report test performance. -(b) Display the top 5 correct predictions and the top 5 incorrect predictions in each class (show the images and the prediction labels) compactly.

Finetune the Resnet18 model's parameters suitably and repeat parts (a) and (b) from above. Compare the performance of finetuning versus using extracted features.