1. Worked on image classification, object detection for checkout free solution
1.1 Built convolutional neural network model to do image classification using PyTorch, trained it on synthetic data of different products generated by Unity and achieved an accuracy of 95%
1.2 Trained object detection model on synthetic data of products. Analyzed failure cases and regenerated training data with new strategies to get better results, got bounding box IoU of 70%
2. Worked on human pose estimation, segmentation and multi-task model
2.1 Built models for pose estimation and segmentation, trained it on datasets combining real images and synthetic images to get a better prediction
2.2 Built multi-task model to predict pose and segmentation at the same time, modified the model to make one task enhance the other