Apply for a GPU community grant: Academic project

#8
by l1aSong - opened
Collov Labs org

Collov-Labs/d-edit is the public demo for D-Edit (https://arxiv.org/abs/2403.04880), an open research project for text-guided image editing.

The Space currently has 84 likes and more than 102,000 lifetime visits. Its optimization and diffusion editing stages require CUDA and cannot complete on CPU Basic. Step 1 segmentation is available on CPU, but the main editing workflow requires GPU hardware.

We request a Community GPU Grant to keep this research demo publicly accessible and reproducible. We have optimized the public configuration for responsible resource usage: the default training batch size is 1, requests run with concurrency 1, the segmentation model is cached, and GPU memory is released between optimization and editing. A T4-class GPU with at least 16 GB VRAM should be sufficient for the public demo.

Paper: https://arxiv.org/abs/2403.04880
Space: https://huggingface.co/spaces/Collov-Labs/d-edit

Sign up or log in to comment