Dec 2023 - Jan 2024
• Developed AlexNet and Resnet-18 CNN models on CIFAR-10, employing research paper architectures enhancedwith image augmentation, regularisation techniques and hyperparametr tuning, for accurate object detection.
• Created a Vision Transformer from scratch trained on MRI Image Dataset by enhancing its performance using various dropout regularisation and hyperparameter tuning and visualized its Attention map in detecting brain tumour patches not detectable by CNN models.
•Developed a conditional-DCGANs from scratch to generate 10,000 unique customised fashion products from random noise image by training on large FashionMNIST Dataset containing 5,00,000 common fashion products.
•Designed a Swin Transformer based scratch vision model using sliding window and patch merging techniques to be computationally efficient than existing vision transformer and attained more accuracy than CNN architecture for object detection, image semantic segmentation.