1. Mumbai Slum Segmentation

    We developed a Deep Learning solution to the problem of slum mapping and monitoring in Mumbai.

    We curated a custom dataset of slums from Google Earth and using data from the Slum Rehabilitation Authority of India.

    Trained an instance segmentation model (Mask R-CNN) on this dataset to identify and segment slums in satellite imagery. Our work will be published in the NIPS 2018 ML4D workshop.

    Project Site PDF Github

  2. SOP-Generator

    Built a LSTM based Statement of Purpose Generator for grad school in PyTorch

    Curated a custom dataset containing diverse Computer Science statement of purpose drafts and achieved a perplexity of 1.17 after training for 200 epochs.


  3. AR Drone Face Follower

    Implemented an IBVS controller (Image based visual servoing) for recognizing and tracking a face using the AR Drone 2.0

    Used OpenFace for Face Recognition + a simple PID for robot control

    PDF GitHub

  4. Mumbai Rains + Delhi Smog twitter visualizations

    Visualizations are lucid, interpretable and beautiful. This project covers the impact of the 2017 Mumbai Rains and Delhi, and Smog visually

    Project Site Github