Parking Lot Application - Deep Learning, Object Detection
- Developed an image analytics application for a parking lot for identification of empty/occupied slots.
- Built a custom dataset of parking lot images of different parking scenarios, trained YOLO v4 for detection and built a backed server with AWS EC2 infrastructure.
Object Detection Using Transformers
- Built a dataset of Balloons class and trained the DETR on “Balloons” class and obtained better and precise detection results compared to other object detection models with an improved testing accuracy of about 3%
Object detection using YOLO on Custom dataset – Deep learning, PyTorch
- Built custom dataset with four classes Hardhat, Vest, Mask, and Boots and annotated it for training and tested the model running it on a video and observed an improvement in validation accuracy by 4%
CIFAR-10 Image Classification using a custom neural network - Deep learning, PyTorch
- Developed and trained a custom neural network by applying regularization and data augmentations on the data and achieved an accuracy of 93.56% over 10 classes
- Also, Adapted One cycle LR finder policy for training to reduce the cost of training and for model assessment at the initial stages of training.