This project involves creating a multi-task model that does classification, localization and segmentaiton using VGG11 backbone and training it from scratch on Oxford-IIIT Pets dataset.
The link to the Github Repository is: Repo
Following is the link for experimentation done during training the model. The results are summarised in this W&B Report: Report
The classes related to models in /models are written manually. The wanb logging scripts are generated by AI
The models are training using ImageNet Weights in the first satge to check if the code written and the setup are working. Once, code is verified, the models are trained from scratch using Xavier initialization and the same models are submitted for evaluation in GradeScope.