Instructions to use facebook/sam3 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use facebook/sam3 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("mask-generation", model="facebook/sam3")# Load model directly from transformers import AutoProcessor, AutoModel processor = AutoProcessor.from_pretrained("facebook/sam3") model = AutoModel.from_pretrained("facebook/sam3", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Request for Manual Review / Reset of Rejected SAM 3 Access — Academic AVTrack Reproduction
Dear SAM 3 Team,
I am a Master's student at the University of Science and Technology Beijing, affiliated with the School of Computer & Communication Engineering. My Hugging Face username is moonlight321.
I am conducting non-commercial academic research on audio-visual person tracking. My current project is to reproduce and extend the official AVTrack pipeline from Fudan University:
Paper: “AVTrack: Audio-Visual Tracking in Human-centric Complex Scenes”
Code: https://github.com/FudanCVL/AVTrack
SAM 3 is necessary for a faithful reproduction because the official AVTrack implementation directly uses Sam3VideoModel and Sam3VideoProcessor with the text prompt “person” to detect and track all human instances across video frames. The resulting boxes and masks are then matched with speaker-localization boxes produced by a vision-language model. Replacing SAM 3 with SAM 2 or another tracker would change this core component and would no longer constitute a strict reproduction of the published baseline.
My previous access request was rejected, possibly because my initial application did not provide enough research context. I respectfully request a manual review and ask whether my access status can be reset to pending or approved.
My intended use is:
- strictly non-commercial academic research;
- reproduction and evaluation of the published AVTrack baseline;
- experiments performed only on our laboratory server;
- no redistribution, sharing, mirroring, or public hosting of the model weights;
- no public inference endpoint based on the restricted weights;
- compliance with the SAM license, Meta policies, and Hugging Face terms.
I am willing to provide my supervisor’s name, institutional webpage, student verification, or further research details through a private channel if required.
Thank you for reviewing my request and for making SAM research available to the academic community.
Sincerely,
moonlight321
University of Science and Technology Beijing
School of Computer & Communication Engineering