Spaces:
Configuration error
Apply for community grant: Academic project (gpu)
I'm building ColorMNet, an open-source video colorization model based on our ECCV 2024 paper.
ColorMNet is an end-to-end trainable framework that leverages memory-based feature propagation, large-scale pretrained visual representations, and local spatiotemporal attention to achieve temporally consistent and vivid video colorization.
This Space aims to make the research easily accessible for the community — allowing users to upload grayscale videos and visualize colorized outputs in real time.
The model requires GPU acceleration (preferably A10 or T4, 30G or more) for efficient inference due to its attention-based temporal modeling.
I'm maintaining this project as part of an open research effort in video restoration and generation, and plan to release all inference code, pretrained weights, and a demo under an open-source license.
GPU support from Hugging Face would help make this work publicly accessible for educational, research, and creative use.
Keywords: video colorization, diffusion-based restoration, open research, temporal consistency, ECCV 2024