Spaces:
Sleeping
Apply for a GPU community grant: Personal project
Note: The application has been fully developed and tested locally with NVIDIA GPU acceleration. The current Hugging Face Space is running on CPU Basic, which cannot execute the complete inference pipeline efficiently, hence this GPU grant request.
Editify is an open-source full-stack web application that converts a static poster into an editable layered design using a computer vision pipeline built from scratch.
The backend integrates GroundingDINO for object detection, SAM2 for pixel-level segmentation, PaddleOCR for text extraction, and LaMa for background reconstruction after removing detected objects. The frontend is built with React and Fabric.js, while the backend uses FastAPI, Docker, and PostgreSQL (Neon) for authentication and project persistence.
The goal of the project is to demonstrate how multiple modern computer vision models can be integrated into a production-style editing workflow. Users can upload a poster, automatically separate objects and text into editable layers, modify them independently, and save their projects through a web interface.
The complete inference pipeline relies on GPU acceleration because GroundingDINO, SAM2, and LaMa are designed for CUDA-based inference. On CPU hardware, processing takes several minutes and the interactive experience is no longer practical. A GPU-enabled Space would allow students, developers, and recruiters to experience the application as intended and explore a real-world end-to-end computer vision system.
GitHub: https://github.com/nishak05/Editify
