/Image-Feature-Extraction-For-Classification

Image Classification, Image Feature Extraction, CNNs, Finetuning, Resnet18, Torchvision, Multi-Class Logistic Regression

Primary LanguageJupyter Notebook

Image Feature Extraction for Image Classification Using CNN Model, Finetuning and Resnet18 model (From torchvision)

  • 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.