Sapiens segmentation (fastforks)

Sapiens (Khirodkar et al. 2024) human body-part segmentation: a ViT pretrained on 300 M human images, fine-tuned with a deconvolution head on Goliath's 28 classes (face, hair, left / right limbs, clothing, lips, teeth, ...); 1024 x 768 input, the Sapiens-Lite TorchScript checkpoints.

This repo is the model card only: it holds no weights. fastforks downloads the authors' weights from the links below and checks them against the SHA-256 in the model card. Code: https://github.com/kleinicke/fastforks. Kernels for the kernels backend: https://huggingface.co/fastforks/kernels. Original code: https://github.com/facebookresearch/sapiens at 2cb07227a740.

Author weight repos: https://huggingface.co/facebook/sapiens-seg-0.3b-torchscript, https://huggingface.co/facebook/sapiens-seg-1b-torchscript.

Task: segmentation. Tier: stable. Outputs: logits. Backends: portable, kernels.

Input: one RGB image. Output: per-pixel class labels (logits with logits=True).

Variants

Weights are the authors' files (not hosted here). SHA-256 shortened to 12 hex digits; the full hashes are in card.toml.

variant	weights (author link)	SHA-256	weights licence	commercial use
sapiens-seg-0.3b-goliath@1	sapiens_0.3b_goliath_best_goliath_mIoU_7673_epoch_194_torchscript.pt2	735a9a8d63fe	CC-BY-NC-4.0	no
sapiens-seg-1b-goliath@1	sapiens_1b_goliath_best_goliath_mIoU_7994_epoch_151_torchscript.pt2	33bba30f3de8	CC-BY-NC-4.0	no

Licence

Original-code strings, recorded as they are (research-use package; check them before any other use). Code: CC-BY-NC-4.0. Files: LICENSE.

Training data: MAE pretraining on Humans-300M (Meta-internal, not released); segmentation fine-tuning on Goliath body-part annotations (Meta, not released).

Licence evidence: https://huggingface.co/facebook/sapiens-seg-1b-torchscript/blob/584504252a6d2a27ea6560927811db734ee42332/README.md

Input size

any H x W >= 32, padded to a multiple of 32 Tested up to (portable, fp16, RTX 5060 Ti 16 GB): sapiens-seg-0.3b-goliath 3000x4000 (2.2 GiB peak allocated, 3.2 reserved); sapiens-seg-1b-goliath 3000x4000 (3.8 GiB peak allocated, 5.0 reserved)

Speed and memory

Segmentation, synthetic, 1024x2048

model	time (ms): original fp32	original	portable	kernels	kernels speed-up: vs original	vs portable	peak memory (MiB, allocated / reserved): original	portable	kernels
sapiens-seg-0.3b-goliath	466	173	132	92.6	1.9x	1.4x	2520 / 2724	1643 / 1762	1643 / 1762
sapiens-seg-1b-goliath	1189	354	275	185	1.9x	1.5x	5721 / 5906	3259 / 3384	3259 / 3382

Startup and first call

model	load (s): original	portable	kernels	first call (s): original	portable	kernels
sapiens-seg-0.3b-goliath	4.92 (512x512)	1.22	0.96†	1.02 (512x512)	0.13†	0.09†
sapiens-seg-1b-goliath	11.33 (512x512)	3.54	2.27†	0.57 (512x512)	0.27†	0.18†

torch.compile and TensorRT

model	input	portable ms	compile ms	compile s (cold)	TensorRT fp16 ms	ONNX + build s	kernels ms	peak MiB (alloc / reserved): portable	compile	TensorRT (device)
sapiens-seg-0.3b-goliath	1024x2048	132	131.9	0.3	156.3	53	92.6	1643 / 1762	1640 / 1762	1564

Fixed-size speed and peak memory, all models

model	mode	480x640	720x1280	1080x1920	12 MP 3000x4000
sapiens-seg-0.3b-goliath	portable	131.3 ms / 1621 / 1960	132.2 ms / 1627 / 1964	133.7 ms / 1641 / 1950	146.0 ms / 2254 / 3302
sapiens-seg-0.3b-goliath	kernels	92.6 ms / 1621 / 1960	93.5 ms / 1627 / 1964	95.2 ms / 1641 / 1950	107.5 ms / 2254 / 3302
sapiens-seg-1b-goliath	portable	274.9 ms / 3238 / 3718	275.7 ms / 3245 / 3734	277.0 ms / 3259 / 3730	289.7 ms / 3869 / 5076
sapiens-seg-1b-goliath	kernels	185.7 ms / 3238 / 3718	186.6 ms / 3245 / 3734	187.6 ms / 3259 / 3730	199.5 ms / 3869 / 5076

Fixed-size cells: time (ms) / peak allocated / peak reserved memory (MiB).

Hardware, protocol and caveats: benchmarks (RTX 5060 Ti 16 GB, sm_120). Original = the authors' code at the precision of portable (fp16, or (fp32) where fp16 is unsupported or not finite); original fp32 = its default fp32 run. Peak memory is the torch allocator's.

Accuracy

Parity with the original code

Every variant is checked against reference outputs of the original code at its pinned commit. CPU, float32: max |diff| <= 0.002 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.05 and p99 <= 0.25 (absolute error of the output logits).

Usage

from fastforks import segmentation

# non-commercial weights: run with FASTFORKS_NONCOMMERCIAL=1 in the environment
model = segmentation.load("sapiens-seg-0.3b-goliath")  # mode="auto": the preferred backend from recorded measurements
result = model("photo.jpg")
result.labels  # [H, W] per-pixel class index
result.backend  # what ran, e.g. Backend(mode='portable', device='cuda:0', dtype='float16')

Install with pip install fastforks (see https://github.com/kleinicke/fastforks). Other variants: sapiens-seg-0.3b-goliath, sapiens-seg-1b-goliath.

Kernels backend

segmentation.load("sapiens-seg-0.3b-goliath", mode="kernels") runs our CUDA kernels (fastbb, cudnn_hacc, seg_up). It needs a prebuilt kernel for your GPU architecture (downloaded from https://huggingface.co/fastforks/kernels on first use) and a recent NVIDIA driver; if there is none it raises BackendUnavailableError with the reason and never falls back silently. mode="auto" picks kernels when it can run and logs the choice.