Is your feature request related to a problem? Please describe.
Video object tracking is computationally expensive, especially when following many objects at once. Current approaches typically process each object in a separate pass, which wastes resources and limits real-time performance. Having an optimized on-device model for efficient multi-object video segmentation would unlock new edge applications in video editing, augmented reality, robotics, and accessibility.
Details of model being requested
- Model name: SAM 3.1 (Meta Segment Anything Model 3.1)
- Source repo link: https://huggingface.co/facebook/sam3.1
- Research paper link [If applicable]: https://arxiv.org/abs/2511.16719
- Model use case: SAM 3.1 is a unified foundation model from Meta for promptable detection, segmentation, and tracking of objects in images and video. It accepts text prompts, image exemplar prompts, and visual prompts (points, boxes, masks), and can exhaustively segment all instances of an open-vocabulary concept specified by a short text phrase.
Additional context for requested model
Object Multiplex eliminates redundant per-object computation, doubling throughput from 16 to 32 frames per second for videos with a medium number of objects and reducing overall GPU resource requirements. These efficiency gains make SAM 3.1 a strong candidate for on-device deployment, where reduced compute and memory demands align well with mobile and embedded hardware constraints.
SAM 3.1 is a drop-in replacement for SAM 3 and is released as open source.
Is your feature request related to a problem? Please describe.
Video object tracking is computationally expensive, especially when following many objects at once. Current approaches typically process each object in a separate pass, which wastes resources and limits real-time performance. Having an optimized on-device model for efficient multi-object video segmentation would unlock new edge applications in video editing, augmented reality, robotics, and accessibility.
Details of model being requested
Additional context for requested model
Object Multiplex eliminates redundant per-object computation, doubling throughput from 16 to 32 frames per second for videos with a medium number of objects and reducing overall GPU resource requirements. These efficiency gains make SAM 3.1 a strong candidate for on-device deployment, where reduced compute and memory demands align well with mobile and embedded hardware constraints.
SAM 3.1 is a drop-in replacement for SAM 3 and is released as open source.