HQ-SAM (fastforks)

Segment Anything in High Quality (Ke et al. 2023): SAM with one more output token and a small head that fuses the final embedding with an early ViT feature, for sharper mask borders and thin structures; SAM's weights are frozen, so prompts work the same. Each prompt gives one mask (SAM's mask plus the HQ mask, upstream's default); multimask_output picks the best-scored of SAM's three candidates as the SAM part. The portable module() is the image side only (embedding and HQ feature).

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/SysCV/sam-hq at e696978d6035.

Author weight repos: https://huggingface.co/lkeab/hq-sam.

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-hq-vit-b@1 sam_hq_vit_b.pth 14a9d662cd6f Apache-2.0 yes
sam-hq-vit-h@1 sam_hq_vit_h.pth a7ac14a08532 Apache-2.0 yes
sam-hq-vit-l@1 sam_hq_vit_l.pth e1a6c385d62b 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: the HQ head is trained on HQSeg-44K (six existing fine-mask datasets, per the paper; their terms were not checked) on top of SAM's frozen weights (SA-1B, under the SA-1B research licence).

Licence evidence: https://huggingface.co/lkeab/hq-sam/blob/09b02a333b37772133eff3997bdba997867374b7/README.md (license: apache-2.0); code: https://github.com/SysCV/sam-hq/blob/e696978d60352dc9a26b12631cd91781502c6546/LICENSE

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-hq-vit-b 1080x1920 (0.4 GiB peak allocated, 0.5 reserved); sam-hq-vit-h 1080x1920 (1.5 GiB peak allocated, 1.9 reserved); sam-hq-vit-l 1080x1920 (0.8 GiB peak allocated, 1.1 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-hq-vit-b 172 103 49.7 36.8 2.8x 1.4x 1704 / 1756 388 / 422 433 / 564
sam-hq-vit-h 738 311 217 144 2.2x 1.5x 4266 / 4318 1500 / 1566 1541 / 1708
sam-hq-vit-l 412 202 118 81.4 2.5x 1.4x 2969 / 3018 846 / 902 889 / 1046

Startup and first call

model load (s): original portable kernels first call (s): original portable kernels
sam-hq-vit-b 0.77 0.27 1.06 0.19 0.41 0.35
sam-hq-vit-h 3.92 1.34 2.29 0.32 0.52 0.47
sam-hq-vit-l 1.98 0.76 1.64 0.20 0.41 0.40

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-hq-vit-b 1024x1024 49.7 ms 39.4 ms 28.0 ms (image encoder only) 36.8 ms 32.8 s 39.0 s 388 / 422 372 / 422 433 / 564 609 / 642
sam-hq-vit-h 1024x1024 217 ms 187 ms 125 ms (image encoder only) 144 ms 42.3 s 69.1 s 1500 / 1566 1476 / 1526 1541 / 1708 2790 / 2818
sam-hq-vit-l 1024x1024 118 ms 97.0 ms 68.3 ms (image encoder only) 81.4 ms 32.9 s 53.3 s 846 / 902 822 / 902 889 / 1046 1482 / 1538

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-hq-vit-b COCO val2017, every 10th image, point prompts, mIoU best of 3 / single mask % 52.19 / 51.05 52.28 / 51.11 52.19 / 51.04 52.19 / 51.04
sam-hq-vit-l COCO val2017, every 10th image, point prompts, mIoU best of 3 / single mask % 56.04 / 52.52 56.00 / 52.53 56.09 / 52.52 56.08 / 52.52
sam-hq-vit-h COCO val2017, every 10th image, point prompts, mIoU best of 3 / single mask % 56.15 / 53.43 56.09 / 53.45 56.16 / 53.43 56.14 / 53.43

Measured on NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks f2992269, 2026-11-10.

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| <= 0.43 % and p99 <= 1.7 % of the reference's median |value| (absolute error of the output pred_masks).

Usage

from fastforks import promptable_segmentation

model = promptable_segmentation.load("sam-hq-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-hq-vit-b, sam-hq-vit-h, sam-hq-vit-l.

Kernels backend

promptable_segmentation.load("sam-hq-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-hq