Skills: Python, PyTorch, GPU-training, Distributed Training, Cloud Computing, GANs, CNNs, Normalization, LaTex.
Description: My group members and I worked on this project throughout our final year at NIT Warangal. We investigated the effect of different normalization techniques within a deep GAN architecture (Layout2Im) that can generate images given a desired layout as bounding boxes. Attentive normalization was found to outperform competing techniques for the considered datasets. Further, the utility of the attentive GAN-generated images is demonstrated for an image classification task by using them as synthetic data during training to alleviate the class imbalance problem.