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 Reconsideration of Rejected SAM 3 Access — Master's Research on UAV Remote Sensing
Dear SAM 3 Team,
I am writing to sincerely ask whether you could reconsider my rejected access request for facebook/sam3, or reset my request so that I can submit it again if any information was incomplete.
I am Wenyi Zhou, a Master's student at Hohai University in China. My research focuses on UAV remote sensing and geospatial object recognition, and one of the main problems I am currently working on is automatic building extraction from high-resolution UAV orthophotos.
My Hugging Face account now includes a verified Hohai University institutional email, and I have also updated my profile to accurately reflect my academic background and research interests.
I am especially interested in SAM 3 because my research involves UAV imagery collected from different locations, seasons, illumination conditions, and building types. A major difficulty in our work is that models trained on one UAV project often perform poorly when applied to imagery from another area.
I would like to evaluate whether SAM 3's open-vocabulary and promptable segmentation capabilities can help with this problem, particularly for:
-text-prompted building extraction using prompts such as “building” or “house”;
-point- and box-prompted segmentation of individual buildings;
-building extraction across UAV datasets acquired in different regions and conditions;
-buildings partially affected by vegetation, shadows, adjacent structures, or different roof materials;
-generating initial building masks that can later be regularized into accurate building footprints and manually checked.
My purpose is academic research and experimentation as a Master's student at Hohai University. I am currently comparing general-purpose foundation models with task-specific remote-sensing segmentation approaches, and SAM 3 is particularly valuable to my research because its text-prompted open-vocabulary capability allows me to study whether buildings can be extracted across previously unseen UAV projects without training a separate model for every location.
The model would be used locally for my research. I will not redistribute or publish the restricted weights, share my credentials, or provide public access to the model.
I understand that access to SAM 3 is granted at the discretion of the repository authors, and I fully respect the review process. I am only asking for reconsideration because I genuinely believe that my research use is closely related to the capabilities SAM 3 was designed to support.
If my previous application did not clearly communicate my academic affiliation or research purpose, I apologize for that. I would be very grateful if you could either reconsider the request or reset it so that I can submit the information again correctly.
I am also happy to provide any additional information needed to verify my student status, university affiliation, or research purpose.
Hugging Face username: ZZW1502
Name: Wenyi Zhou
Affiliation: Hohai University
Position: Master's Student
Research area: UAV Remote Sensing and Geospatial Object Recognition
Country: China
Institutional email: Verified on my Hugging Face account
Thank you very much for taking the time to read my request and for your work on SAM 3.
Best regards,
Wenyi Zhou