🛰️ Drishti — explainable disaster detection from drone images

Upload an aerial / drone photo and the model labels it as collapsed building, fire / smoke, flood, traffic accident or normal, shows where it looked (Grad-CAM), suggests a response, and routes uncertain frames to a human analyst. The Area scan tab tiles a large frame and maps where the emergency is.

• Model: ImageNet-pretrained CNN (MobileNetV3-Large / EfficientNet-B0, best chosen on validation) fine-tuned in PyTorch, calibrated with temperature scaling.
• Data: AIDER — C. Kyrkou & T. Theocharides, EmergencyNet, IEEE JSTARS 13 (2020) 1687–1699; Zenodo DOI 10.5281/zenodo.3888300 (CC BY 4.0).
• Built for EC-422 Artificial Intelligence, IIT (BHU) Varanasi. Code and training notebook: see the GitHub repository linked in the project README.

This is a decision-support demo trained on web-collected images; it is not a certified emergency system.