Spaces:
Running on Zero
Apply for a GPU community grant: Academic project
PanoSAMic is a panoramic (360°) semantic segmentation model, accepted at ICPR 2026 (DFKI / RPTU Kaiserslautern-Landau). It adapts the frozen SAM ViT encoder for equirectangular images and reaches SotA on Stanford2D3DS and Matterport3D.
Paper: https://huggingface.co/papers/2601.07447
Models: https://huggingface.co/dfki-av/PanoSAMic
GitHub: https://github.com/dfki-av/PanoSAMic
The demo runs the ViT-H backbone, which is too heavy for the free CPU tier — it currently works but is slow. A ZeroGPU grant would make it interactive. Weights and code are already public.
Hi @mahdichamseddine , we've assigned ZeroGPU to this Space. Please check the compatibility and usage sections of the ZeroGPU documentation to make sure your Space runs correctly on ZeroGPU. If you're using a coding agent like Claude Code, you can also try Hugging Face's official huggingface-zerogpu skill, which guides the agent through ZeroGPU's constraints and migration steps.
If you're able to, please consider upgrading to Pro ($9/month) for a higher ZeroGPU quota plus features like Dev Mode and Private Storage.
Thank you, @hysts , much appreciated! I'll go through the ZeroGPU compatibility and usage docs and adapt the Space accordingly. Thanks for the quick turnaround and for the pointers.