Segment Anything 2.1 (SAM 2.1) (fastforks)

Segment Anything 2.1 (Ravi et al. 2024), image mode: promptable segmentation with a hierarchical Hiera image encoder, an FPN neck and the SAM prompt encoder / mask decoder with high-resolution skip features; masks from points, boxes and mask logits. video follows objects prompted on the first frame through a video with the original memory attention / memory encoder (SAM2VideoPredictor, forward, memory features in bfloat16 as upstream); the memory side has no fused kernels yet. Option samurai tracks as SAMURAI (Yang et al. 2024: a Kalman filter on the mask box picks among the multimask outputs, and the memory keeps only confident frames); upstream SAMURAI tracks one object, here every object has its own filter and memory. 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/sam2 at 2b90b9f.

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
sam2.1-hiera-base-plus@1 sam2.1_hiera_base_plus.pt a2345aede871 Apache-2.0 yes
sam2.1-hiera-large@1 sam2.1_hiera_large.pt 2647878d5dfa Apache-2.0 yes
sam2.1-hiera-small@1 sam2.1_hiera_small.pt 6d1aa6f30de5 Apache-2.0 yes
sam2.1-hiera-tiny@1 sam2.1_hiera_tiny.pt 7402e0d864fa 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 and the SA-V video dataset (plus internal data) per the paper; SA-V itself is under CC BY 4.0, SA-1B under its research licence.

Licence evidence: https://github.com/facebookresearch/sam2/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): sam2.1-hiera-base-plus 1080x1920 (0.4 GiB peak allocated, 0.6 reserved); sam2.1-hiera-large 1080x1920 (0.7 GiB peak allocated, 0.9 reserved); sam2.1-hiera-small 1080x1920 (0.3 GiB peak allocated, 0.4 reserved); sam2.1-hiera-tiny 1080x1920 (0.3 GiB peak allocated, 0.4 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
sam2.1-hiera-base-plus 107 47.2 32.0 25.1 1.9x 1.3x 823 / 870 412 / 464 426 / 524
sam2.1-hiera-large 242 86.4 68.0 48.8 1.8x 1.4x 1656 / 1724 742 / 768 760 / 828
sam2.1-hiera-small 74.8 41.8 18.8 15.6 2.7x 1.2x 618 / 662 318 / 366 330 / 386
sam2.1-hiera-tiny 66.3 38.0 16.6 13.9 2.7x 1.2x 577 / 636 304 / 382 316 / 362

Startup and first call

model load (s): original portable kernels first call (s): original portable kernels
sam2.1-hiera-base-plus 0.74 0.43 0.19โ€  0.14 0.03โ€  0.03โ€ 
sam2.1-hiera-large 1.95 1.10 0.39โ€  0.18 0.07โ€  0.05โ€ 
sam2.1-hiera-small 0.48 0.36 0.13โ€  0.14 0.02โ€  0.02โ€ 
sam2.1-hiera-tiny 0.41 0.25 0.11โ€  0.15 0.02โ€  0.01โ€ 

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)
sam2.1-hiera-base-plus 1024x1024 32.0 ms 26.4 ms 16.7 ms (image encoder only) 25.1 ms 28.3 s 43.7 s 412 / 464 365 / 404 426 / 524 588 / 620
sam2.1-hiera-large 1024x1024 68.0 ms 59.5 ms 38.1 ms (image encoder only) 48.8 ms 38.9 s 55.3 s 742 / 768 678 / 748 760 / 828 1212 / 1232
sam2.1-hiera-small 1024x1024 18.8 ms 14.9 ms 9.93 ms (image encoder only) 15.6 ms 24.5 s 40.2 s 318 / 366 283 / 326 330 / 386 429 / 452
sam2.1-hiera-tiny 1024x1024 16.6 ms 14.2 ms 8.56 ms (image encoder only) 13.9 ms 13.4 s 34.1 s 304 / 382 270 / 322 316 / 362 414 / 446

Fixed-size speed and peak memory, all models

model mode 480x640 720x1280 1080x1920 12 MP 3000x4000
sam2.1-hiera-base-plus portable 32.0 ms / 395 / 540 31.9 ms / 402 / 558 32.1 ms / 415 / 568 -
sam2.1-hiera-base-plus kernels 25.1 ms / 408 / 472 25.1 ms / 415 / 490 25.3 ms / 428 / 472 -
sam2.1-hiera-large portable 68.2 ms / 724 / 936 68.6 ms / 732 / 948 68.3 ms / 745 / 954 -
sam2.1-hiera-large kernels 48.7 ms / 742 / 824 48.7 ms / 750 / 836 49.0 ms / 763 / 842 -
sam2.1-hiera-small portable 18.7 ms / 298 / 420 19.1 ms / 305 / 422 18.9 ms / 319 / 426 -
sam2.1-hiera-small kernels 15.4 ms / 310 / 356 15.5 ms / 317 / 358 15.6 ms / 331 / 346 -
sam2.1-hiera-tiny portable 16.7 ms / 285 / 410 16.6 ms / 292 / 410 16.8 ms / 305 / 426 -
sam2.1-hiera-tiny kernels 13.8 ms / 296 / 346 13.8 ms / 304 / 346 14.0 ms / 317 / 346 -

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. Semi-supervised: every object's first-frame mask as input_masks, video forward at 480p; the lower object id wins where masks overlap; J (IoU) and F (boundary) as davis2017-evaluation, mean over objects. Original = the original code named under Details, same checkpoint and protocol.

variant dataset original fp32 original fp16 portable kernels note
sam2.1-hiera-tiny COCO val2017, every 10th image, point prompts, mIoU best of 3 / single mask % 57.63 / 57.69 57.56 / 57.64 57.62 / 57.72 57.61 / 57.77
sam2.1-hiera-tiny DAVIS 2017 val, every 3rd sequence, video tracking, J&F / J / F % 83.80 / 79.70 / 87.90 83.86 / 79.50 / 88.22 83.76 / 79.68 / 87.84 83.77 / 79.68 / 87.85
sam2.1-hiera-tiny DAVIS 2017 val, video tracking, J&F / J / F % 87.54 / 84.83 / 90.26 87.65 / 84.86 / 90.45 87.53 / 84.84 / 90.23 87.54 / 84.84 / 90.24
sam2.1-hiera-small COCO val2017, every 10th image, point prompts, mIoU best of 3 / single mask % 59.39 / 58.46 59.34 / 58.42 59.38 / 58.49 59.38 / 58.47
sam2.1-hiera-small DAVIS 2017 val, every 3rd sequence, video tracking, J&F / J / F % 84.08 / 80.26 / 87.90 84.11 / 80.28 / 87.93 84.13 / 80.31 / 87.94 84.13 / 80.31 / 87.95
sam2.1-hiera-base-plus COCO val2017, every 10th image, point prompts, mIoU best of 3 / single mask % 58.79 / 58.67 58.73 / 58.57 58.81 / 58.71 58.82 / 58.73
sam2.1-hiera-base-plus DAVIS 2017 val, every 3rd sequence, video tracking, J&F / J / F % 83.88 / 80.62 / 87.14 83.88 / 80.61 / 87.16 83.87 / 80.61 / 87.13 83.86 / 80.60 / 87.12
sam2.1-hiera-large COCO val2017, every 10th image, point prompts, mIoU best of 3 / single mask % 64.88 / 64.71 64.77 / 64.73 64.86 / 64.68 64.87 / 64.70
sam2.1-hiera-large DAVIS 2017 val, every 3rd sequence, video tracking, J&F / J / F % 85.91 / 82.21 / 89.62 85.98 / 82.25 / 89.71 85.92 / 82.22 / 89.63 85.92 / 82.22 / 89.63

Measured on NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 84ba91d4 / 939adfe8, 2026-11-06 to 2026-11-12.

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("sam2.1-hiera-base-plus")  # 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: sam2.1-hiera-base-plus, sam2.1-hiera-large, sam2.1-hiera-small, sam2.1-hiera-tiny.

Kernels backend

promptable_segmentation.load("sam2.1-hiera-base-plus", 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/sam2