AI_OF_GOD

---

• This is a College Technical Fest Concetto 2023's Competition in which we used HiRISE Dataset that contains 8 different classes of mars surface image. We were the first in the competition.
• The main challenge was that the given dataset is so imbalanced that getting a great accuracy is critical.
• We did Image Classification on this dataset using Deep Learning, we used different models such as ShuffleNetV2, EfficientNet, MobileNet and also developed a Custom CNN Model from scratch using Python and Pytorch.
• We built Dataloader, Pre-processing Dataset, Data Augmentation, Warm Epochs, Learning Scheduler from scratch using Python and Pytorch.
• We also implemented and used different losses for Image Classification such as Categorical Cross-entropy, Cosine Loss, Focal Cosine Loss, Focal Loss from different research papers to get better accuracy.
• We also built my own code to store my model's weights for each epoch.
• We also created a code so you can see your results of each epoch stored in a Google Doc that you can access through your phone from anywhere in the world.