facebook/sapiens2-matting-1b

Sapiens2-1B-Matting

Per-pixel human image matting with a soft alpha matte and pre-multiplied foreground RGB output.

This repository contains the 1B Human Matting checkpoint, finetuned from the Sapiens2-1B pretrained backbone.

• 📄 Paper: arXiv:2604.21681
• 🌐 Project Page: rawalkhirodkar.github.io/sapiens2
• 💻 Code: github.com/facebookresearch/sapiens2

Model Details

• Developed by: Meta
• Model type: Vision Transformer
• License: Sapiens2 License
• Task: Human image matting
• Base model: facebook/sapiens2-pretrain-1b
• Format: safetensors
• File: sapiens2_1b_matting.safetensors

Quick Start

Install the Sapiens2 repo (pip install -e .), download the checkpoint, and run the demo:

# 1. Download the checkpoint to $SAPIENS_CHECKPOINT_ROOT/matting/
hf download facebook/sapiens2-matting-1b sapiens2_1b_matting.safetensors \
    --local-dir ~/sapiens2_host/matting

# 2. Run the demo (edit INPUT, OUTPUT, and MODEL_NAME inside the script)
cd $SAPIENS_ROOT/sapiens/dense
./scripts/demo/matting.sh

See the Human Matting guide for details on inputs, outputs, and visualization options.

Model Card

Field	Value
Architecture	Sapiens2 ViT backbone + Human Matting head
Backbone parameters	1.462 B
Backbone FLOPs	4.715 T
Embedding dim	1536
Layers	40
Attention heads	24
Inference resolution	1024 × 768 (H × W)
Patch size	16

Sapiens2-Matting Family

Model	Params	FLOPs	Embed dim	Layers	Heads
Sapiens2-1B (this)	1.462 B	4.715 T	1536	40	24

See the Sapiens2 Collection for all variants and other downstream task checkpoints.

Intended Use

• Human image matting on human-centric imagery
• Research on human-centric vision

License

Released under the Sapiens2 License.

Citation

@article{khirodkarsapiens2,
  title={Sapiens2},
  author={Khirodkar, Rawal and Wen, He and Martinez, Julieta and Dong, Yuan and Su, Zhaoen and Saito, Shunsuke},
  journal={arXiv preprint arXiv:2604.21681},
  year={2026}
}