|
| Prompts: person, bicycle, dog, yellow car |
|---|
| 
|
| File | Size | Mask IoU vs BF16 (mean / min) |
|---|---|---|
| `model.safetensors` (BF16) | 1.68 GB | 1.0000 / 1.0000 |
| `sam3.1-INT8-ConvRot.safetensors` | 1.05 GB | 0.9984 / 0.9913 |
| `sam3.1-INT8.safetensors` | 0.99 GB | 0.9976 / 0.9829 |
| `sam3.1-FP8.safetensors` | 1.05 GB | 0.9971 / 0.9805 |
53 text prompts on 6 images; 1.0 means identical masks. Unsloth Studio uses INT8-ConvRot on NVIDIA GPUs and BF16 elsewhere.
## Usage
```python
import torch
from PIL import Image
from transformers import Sam3Model, Sam3Processor
model = Sam3Model.from_pretrained("unsloth/sam3.1", dtype=torch.bfloat16).to("cuda").eval()
processor = Sam3Processor.from_pretrained("unsloth/sam3.1")
image = Image.open("living_room.png").convert("RGB")
inputs = processor(images=image, text="sofa", return_tensors="pt").to("cuda", torch.bfloat16)
with torch.inference_mode():
outputs = model(**inputs)
results = processor.post_process_instance_segmentation(
outputs, threshold=0.5, mask_threshold=0.5, target_sizes=inputs.get("original_sizes").tolist()
)[0]
print(results["scores"], results["boxes"], results["masks"].shape)
```
---
# SAM 3.1
[SAM 3](https://github.com/facebookresearch/sam3) (Segment Anything with Concepts) is a unified foundation model from Meta for promptable segmentation in images and videos. It detects, segments, and tracks objects using text or visual prompts such as points, boxes, and masks. SAM 3 introduces the ability to exhaustively segment all instances of an open-vocabulary concept specified by a short text phrase, handling over 50x more unique concepts than existing benchmarks. SAM 3.1 builds on this with **Object Multiplex**, a shared-memory approach for joint multi-object tracking that delivers ~7x faster inference at 128 objects on a single H100 GPU without sacrificing accuracy, along with improved VOS performance on 6 out of 7 benchmarks.
