Segment Anything (SAM) (fastforks)

Segment Anything (Kirillov et al. 2023): promptable segmentation with a ViT image encoder (windowed + global attention, relative positions) and a light prompt encoder / mask decoder; masks from points, boxes and mask logits. The portable module() is the image encoder only.

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/segment-anything at dca509f.

Task: promptable_segmentation. Tier: stable. Outputs: pred_masks, iou_scores. Backends: portable, kernels.

Input: one image plus a point, box or text prompt. Output: masks with quality scores.

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
sam-vit-b@1 sam_vit_b_01ec64.pth ec2df6273261 Apache-2.0 yes
sam-vit-h@1 sam_vit_h_4b8939.pth a7bf3b02f3eb Apache-2.0 yes
sam-vit-l@1 sam_vit_l_0b3195.pth 3adcc4315b64 Apache-2.0 yes

Licence

Original-code strings, recorded as they are (research-use package; check them before any other use). Code: Apache-2.0. Licence files: LICENSE.

Training data: SA-1B (11M licensed images, 1.1B automatically generated masks) per the paper; the dataset itself is under the SA-1B research licence.

Licence evidence: https://github.com/facebookresearch/segment-anything/blob/main/README.md (License section: model checkpoints and code under Apache 2.0)

Input size

any H x W >= 1, resized inside to the fixed encoder input 1024x1024 (the default export size) Tested up to (portable, fp16, RTX 5060 Ti 16 GB): sam-vit-b 1080x1920 (0.4 GiB peak allocated, 0.5 reserved); sam-vit-h 1080x1920 (1.5 GiB peak allocated, 1.9 reserved); sam-vit-l 1080x1920 (0.8 GiB peak allocated, 1.0 reserved)

Speed and memory

Promptable segmentation, synthetic, 1024x1024

model time (ms): original fp32 original portable kernels kernels speed-up: vs original vs portable peak memory (MiB, allocated / reserved): original portable kernels
sam-vit-b 168 98.2 48.3 37.0 2.7x 1.3x 1769 / 1802 378 / 420 377 / 420
sam-vit-h 734 306 214 143 2.1x 1.5x 5295 / 5404 1484 / 2524 1480 / 1524
sam-vit-l 406 197 116 81.5 2.4x 1.4x 3411 / 3464 833 / 1262 831 / 862

Startup and first call

model load (s): original portable kernels first call (s): original portable kernels
sam-vit-b 0.75 0.46 0.17โ€  0.18 0.05โ€  0.04โ€ 
sam-vit-h 4.61 2.45 0.89โ€  0.38 0.21โ€  0.14โ€ 
sam-vit-l 1.80 1.21 0.43โ€  0.27 0.12โ€  0.08โ€ 

torch.compile and TensorRT

model input time: portable compile TensorRT fp16 kernels preparation: compile (cold first call) ONNX + engine build peak MiB (alloc / reserved): portable compile kernels TensorRT (engine + torch)
sam-vit-b 1024x1024 48.3 ms 40.9 ms 27.3 ms (image encoder only) 37.0 ms 14.5 s 36.4 s 378 / 420 362 / 400 377 / 420 545 / 582
sam-vit-h 1024x1024 214 ms 188 ms 124 ms (image encoder only) 143 ms 42.6 s 69.7 s 1484 / 2524 1436 / 1484 1480 / 1524 2744 / 2784
sam-vit-l 1024x1024 116 ms 97.2 ms 67.5 ms (image encoder only) 81.5 ms 32.9 s 53.4 s 833 / 1262 793 / 862 831 / 862 1453 / 1520

Fixed-size speed and peak memory, all models

model mode 480x640 720x1280 1080x1920 12 MP 3000x4000
sam-vit-b portable 48.6 ms / 367 / 482 48.6 ms / 374 / 482 49.0 ms / 387 / 484 -
sam-vit-b kernels 37.3 ms / 364 / 424 37.4 ms / 370 / 404 38.1 ms / 384 / 432 -
sam-vit-h portable 216.1 ms / 1490 / 1882 215.4 ms / 1496 / 1902 215.7 ms / 1510 / 1916 -
sam-vit-h kernels 143.6 ms / 1459 / 1712 144.4 ms / 1466 / 1704 145.1 ms / 1479 / 1714 -
sam-vit-l portable 117.2 ms / 814 / 1052 116.5 ms / 820 / 1060 117.5 ms / 834 / 1048 -
sam-vit-l kernels 82.0 ms / 812 / 930 81.6 ms / 818 / 946 83.2 ms / 833 / 922 -

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 original code named above 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

On public datasets

One point per ground-truth instance (the interior point farthest from the border); IoU of the best-scored of 3 masks / of the single mask. Original = the original code named under Details, same checkpoint and protocol.

variant dataset original fp32 original fp16 portable kernels note
sam-vit-b COCO val2017, every 10th image, point prompts, mIoU best of 3 / single mask % 52.91 / 50.31 53.08 / 50.39 52.91 / 50.31 52.89 / 50.32
sam-vit-l COCO val2017, every 10th image, point prompts, mIoU best of 3 / single mask % 57.89 / 53.32 57.76 / 53.31 57.84 / 53.32 57.85 / 53.33
sam-vit-h COCO val2017, every 10th image, point prompts, mIoU best of 3 / single mask % 57.93 / 53.13 57.92 / 53.13 57.97 / 53.12 57.99 / 53.12

Measured on NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 939adfe8, 2026-11-06.

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| <= 3 % and p99 <= 15 % of the reference's median |value| (absolute error of the output pred_masks).

Usage

from fastforks import promptable_segmentation

model = promptable_segmentation.load("sam-vit-b")  # mode="auto": the preferred backend from recorded measurements
result = model("photo.jpg", input_points=[[x, y]], input_labels=[1])
result.masks  # [M, H, W] bool
# SAM 3 (text prompts): model("photo.jpg", text="coin") -> boxes, scores, masks
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: sam-vit-b, sam-vit-h, sam-vit-l.

Kernels backend

promptable_segmentation.load("sam-vit-b", mode="kernels") runs our CUDA kernels (dpt_fused). 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.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Collection including fastforks/sam