Thanks! I’m using a ResNet50 model pre-trained on ImageNet, fine-tuned on a custom Mars terrain dataset of 5,000 images (crater, dusty, rocky, sandy). The training is done with PyTorch, using data augmentation (rotation, flip, color jitter) to improve generalization. Validation split is 80/20. Planning to add real-time prediction via webcam and possibly deploy as a simple web app
Comments 3
wow, super cool - how are you planning to implement this?
Thanks! I’m using a ResNet50 model pre-trained on ImageNet, fine-tuned on a custom Mars terrain dataset of 5,000 images (crater, dusty, rocky, sandy). The training is done with PyTorch, using data augmentation (rotation, flip, color jitter) to improve generalization. Validation split is 80/20. Planning to add real-time prediction via webcam and possibly deploy as a simple web app
but I might change the categories later - having some overlap issues. Using synthetic data for now, real NASA dataset was too big (50GB+)
Sign in to join the conversation.