facebook/sapiens

Model Details

Sapiens, a family of models for four fundamental human-centric vision tasks - 2D pose estimation, body-part segmentation, depth estimation, and surface normal prediction. Our models natively support 1K high-resolution inference and are extremely easy to adapt for individual tasks by simply fine-tuning models pretrained on over 300 million in-the-wild human images. The resulting models exhibit remarkable generalization to in-the-wild data, even when labeled data is scarce or entirely synthetic. Our simple model design also brings scalability - model performance across tasks improves as we scale the parameters from 0.3 to 2 billion. Sapiens consistently surpasses existing baselines across various human-centric benchmarks.

Model Description

• Developed by: Meta
• Model type: Vision Transformers
• License: Creative Commons Attribution-NonCommercial 4.0

More Resources

• Repository: https://github.com/facebookresearch/sapiens
• Paper: https://arxiv.org/abs/2408.12569
• Demos: Sapiens Gradio Spaces
• Project Page: https://about.meta.com/realitylabs/codecavatars/sapiens
• Additional Results: https://rawalkhirodkar.github.io/sapiens
• HuggingFace Collection: https://huggingface.co/collections/facebook/sapiens-66d22047daa6402d565cb2fc

Uses

• pose estimation (keypoints 17, keypoints 133, keypoints 308)
• body-part segmentation (28 classes)
• depth estimation
• surface normal estimation

Model Zoo

Note: This repository does not host any checkpoints but contains links to all the model repositories.

We provide checkpoints in three formats:

• original: weights can be finetuned for your use case along with inference.
• torchscript: (inference only) weights ported to torchscript.
• bfloat16: (inference only) for large scale processing, weights ported to bfloat16 (A100 gpu only + pytorch-2.3).

Model Name	Original	TorchScript	BFloat16
sapiens-pretrain-0.3b	link	link	link
sapiens-pretrain-0.6b	link	link	link
sapiens-pretrain-1b	link	link	link
sapiens-pretrain-2b	link	link	link
sapiens-pose-0.3b	link	link	link
sapiens-pose-0.6b	link	link	link
sapiens-pose-1b	link	link	link
sapiens-seg-0.3b	link	link	link
sapiens-seg-0.6b	link	link	link
sapiens-seg-1b	link	link	link
sapiens-depth-0.3b	link	link	link
sapiens-depth-0.6b	link	link	link
sapiens-depth-1b	link	link	link
sapiens-depth-2b	link	link	link
sapiens-normal-0.3b	link	link	link
sapiens-normal-0.6b	link	link	link
sapiens-normal-1b	link	link	link
sapiens-normal-2b	link	link	link

Helper models for bounding box detection or background removal.

Model Name	Original	TorchScript	BFloat16
sapiens-pose-bbox-detector	link	-	-
sapiens-seg-foreground-1b	-	link	-

Other finetuned models (pose-133 and pose-17): here