Models · Edges and lines

HED soft edges (ControlNet 1.1)

HED edges (Xie and Tu, ICCV 2015; ControlNet's Apache-2.0 re-implementation): soft, thick edge maps of a photo at several scales, the input of the soft-edge ControlNet. Apache-2.0 code and weights.

experimental Fork of huggingface/controlnet_aux@b1aac31 · backends: portable, kernels

Checkpoint hed

Paper 2015 Added in 0.2.0

Portable, RTX 5060 Ti 16 GB 11.3 ms 1080x1920, fp16
vs the faster original 3.4× original fp16: 38.5 ms
Peak memory 187 MiB original: 300 MiB
Accuracy vs original fp32 – no test-set run

Use it

pip install fastforks

Lean install with only what this model needs: pip install fastforks-core

from fastforks import edges

model = edges.load("hed")  # kernels on a supported GPU, else portable PyTorch
result = model(image)
result.edges
result.backend  # what ran: mode, device, dtype, kernels
fastforks bench hed

Times the model on this machine. load(…, mode="portable") or mode="kernels" picks a backend; usage guide.

Speed and memory

Every backend on one GPU (RTX 5060 Ti 16 GB) and the same input, the authors’ code included. Kernels runs fp16. Portable runs fp16 on CUDA.

BackenddtypeTime per call (ms)vs original fp16Peak memory
allocated / reserved
Load (s)First call (s)
Original fp32fp32not measured ––––
Original fp16torch.autocast(float16) around the call; huggingface/controlnet_aux@b1aac318b15c hed, host OpenCV resize, float post-processingfp16 38.5–300 MiB/324 MiB0.20.05
Portable fp32fp32not measured ––––
Portablefp16 11.33.4×187 MiB/224 MiB0.20.01
Kernelsnot measured ––––
Portable + torch.compilenot measured here ––––
TensorRTnot measured here ––––

Measured on RTX 5060 Ti 16 GB, torch 2.14.1+cu130, 2026-11-14. First call, 3 warm-up, peak reset, median of 3 rounds x median of 5. Ratios divide two times of this table (same GPU, same input). Original fp16 is the authors’ own precision option, else torch.autocast. torch.compile and TensorRT: build time instead of load time. TensorRT memory (engine, not comparable) is the GPU memory the engine took, read from the driver because TensorRT allocates outside PyTorch, so it can't be compared with the PyTorch peaks of the other rows.

Other devices

DeviceBackenddtypems / callPeak memoryOutput vs referenceInput
Apple M1 Pro 16 GB MPS, torch 2.14.1, 2026-11-14Originalfp3266.8 1,554 MiB process–1024x1024
Portablefp1656.9 1,381 MiB processmatches1024x1024
Portablefp3264.9 1,415 MiB processmatches1024x1024
Kernelsfp32not implemented–1024x1024

CUDA memory is torch’s peak allocated / reserved; Apple has no such counter, so the Mac rows show the process footprint (weights, Python and torch included). “Output vs reference” compares the device’s output with the stored golden output under the card’s tolerance; – = not checked on that device. Kernels “not implemented” = no kernels for this model on that backend yet: it runs portable there.

Licence and weights

Commercial use allowed

Code
Apache-2.0 LICENSE
Weights
Apache-2.0 source
Training data
Not stated for these checkpoints (lllyasviel/Annotators, ControlNet 1.1); HED (Xie and Tu) trains on BSDS500; this re-trained model states no data.

Our reading of the licences, not legal advice.

Weights

ControlNetHED.pth 29.4 MB· hosted on Hugging Face

fastforks downloads and checks these files on first use (fastforks download hed).

Accuracy

The test set through the authors’ code and through ours: same checkpoint, same protocol. The distance between the original’s fp32 and fp16 runs is what half precision alone costs; our runs should stay within it.

Accuracy not measured for hed.

Closeness to the original

Beyond the test-set scores: the raw output of ours against the original’s fp32 output on the same input, pixel by pixel. CPU, float32: max |diff| <= 0.002 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.02 and p99 <= 0.25 (absolute error of the output edges).

CheckToleranceMeasured
CPU fp32, max |diff|≤ 0.002 not recorded
GPU default dtypes, mean |diff|≤ 0.02 0.00006
GPU default dtypes, p99 |diff|≤ 0.25 0.00102

Recorded for hed on RTX 5060 Ti, torch 2.10.0+cu130, 2026-11-17, with these kernels: cudnn_hacccutlass_convigev

Details

Soft edge preprocessor of ControlNet 1.1 (controlnet_aux `HEDdetector`, ComfyUI's "HED Soft-Edge Lines"): lllyasviel's Apache-2.0 re-implementation of holistically-nested edge detection, a VGG-style network whose five side outputs are averaged into a soft edge map (RGB input, smoother edges than the original HED).

Input / output
Input: one RGB image. Output: a line / edge map [H, W] in [0, 1] at the input size (1 = line), the ControlNet preprocessors (line art, HED, PiDiNet, TEED / Anyline); M-LSD also returns straight line segments.
Input size
any H x W >= 32, padded to a multiple of 32; tested up to 1080x1920 (187 MiB allocated)
Options
  • short_side=512>= 64· changes the output
  • safe=false· changes the output
Outputs
edges
CUDA kernels
  • igevIGEV stereo: convolution epilogues, correlation lookups, GRU gates
  • cutlass_convCUTLASS convolutions with the recurrent update (GRU) fused in
  • cudnn_haccconvolutions with fp16 accumulation (cuDNN)
Apple (MPS)
fp32, fp16
Full reference, generated from the repository

HED edges (Xie and Tu, ICCV 2015; ControlNet's Apache-2.0 re-implementation): soft, thick edge maps of a photo at several scales, the input of the soft-edge ControlNet. Apache-2.0 code and weights.

Commercial use: yes.

variantinputbackendtime (ms)peak memory (MiB, allocated / reserved)
hedsynthetic (1080x1920)original
portable
kernels
38.5
11.3
–
300 / 324
187 / 224
–
hedsynthetic (1024x1024)original
portable
kernels
portable + torch.compile
TensorRT fp16
20.3
6.02
–
4.81
2.57
199 / 226
117 / 146
–
137 / 166
170 / 178
Details
  • Original code: https://github.com/huggingface/controlnet_aux at b1aac318b15c
  • Code licence: Apache-2.0 (permissive); files: LICENSE
  • Weights licence: Apache-2.0
  • Licence evidence: https://github.com/huggingface/controlnet_aux/blob/b1aac318b15c4e65a21bc516dc236ad4bf593d46/src/controlnet_aux/hed/__init__.py ; https://huggingface.co/lllyasviel/Annotators
  • Training data: Not stated for these checkpoints (lllyasviel/Annotators, ControlNet 1.1); HED (Xie and Tu) trains on BSDS500; this re-trained model states no data.
  • Options: short_side = 512 (>= 64; also resolution; also detect_resolution), safe = False
  • Kernel libraries: igev, cutlass_conv, cudnn_hacc
  • Variants: hed
  • Outputs: edges
  • Input size: any H x W >= 32, padded to a multiple of 32
  • Tested up to (portable, fp16, RTX 5060 Ti 16 GB): hed 1080x1920 (0.2 GiB peak allocated, 0.2 reserved)
  • hed 1024x1024, first call: portable + torch.compile 9 s (cold Inductor cache), TensorRT fp16 12 s (ONNX + engine).
  • hed original 1080x1920, original 1024x1024: torch.autocast(float16) around the call; huggingface/controlnet_aux@b1aac318b15c hed, host OpenCV resize, float post-processing
  • hed load / first call: original 0.24 s / 0.057 s, portable 0.18 s / 0.012 s
  • Protocol, first call, 3 warm-up, peak reset, median of 3 rounds x median of 5

Parity

CPU, float32: max |diff| <= 0.002 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.02 and p99 <= 0.25 (absolute error of the output edges).

variantkernels error vs original, mean / p99GPU, torchkernels used
hed6e-05 / 0.00102RTX 5060 Ti, torch 2.10.0+cu130cudnn_hacc@cudnn91501, cutlass_conv@988138e042cb:sm_120, igev@f075827d68c6:sm_120

Example output

No example rendered for this checkpoint yet.