Models · Edges and lines

PiDiNet soft edges

PiDiNet (Su et al., ICCV 2021): a small, fast edge detector with pixel-difference convolutions, folded into plain convolutions here; soft edges for the soft-edge ControlNet. Research use only (non-commercial).

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

Checkpoint pidinet

Paper 2021 Added in 0.2.0

Portable, RTX 5060 Ti 16 GB 13.1 ms 1080x1920, fp16
vs the faster original 2.7× original fp16: 35.9 ms
Peak memory 328 MiB original: 339 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[nc]"

from fastforks import edges

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

The weights are non-commercial: fastforks downloads them only with FASTFORKS_NONCOMMERCIAL=1 set.

fastforks bench pidinet

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 pidinet, host OpenCV resize, float post-processingfp16 35.9–339 MiB/368 MiB0.10.06
Portable fp32fp32not measured ––––
Portablefp16 13.12.7×328 MiB/372 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, 2026-11-14Originalfp32109 1,480 MiB process–1024x1024
Portablefp1673.2 1,349 MiB processmatches1024x1024
Portablefp3290.8 1,359 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

Non-commercial only

Code
LicenseRef-research-only LICENSE
Weights
LicenseRef-research-only (assumed from code)(not stated by the authors; assumed to follow the code) source
Training data
Not stated for these checkpoints (lllyasviel/Annotators, ControlNet 1.1). The file name matches the paper's table 5 BSDS500 model.
  • code: LicenseRef-research-only (noncommercial)
  • weights: LicenseRef-research-only (assumed from code)
  • weights licence not stated by the original authors; assumed to be the code licence

Our reading of the licences, not legal advice.

Weights

table5_pidinet.pth 2.9 MB· hosted on Hugging Face

fastforks downloads and checks these files on first use (fastforks download pidinet). Non-commercial weights need FASTFORKS_NONCOMMERCIAL=1.

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 pidinet.

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 0.00019
GPU default dtypes, p99 |diff|≤ 0.25 0.00192

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

Details

Soft edge preprocessor of ControlNet 1.1 (controlnet_aux `PidiNetDetector`, ComfyUI's "PiDiNet Soft-Edge Lines"): pixel difference networks (Su et al., ICCV 2021), a 0.7 M-parameter edge detector of difference convolutions, here converted to plain depthwise convolutions (the authors' re-parameterization). Research use only.

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 (328 MiB allocated)
Options
  • short_side=512>= 64· changes the output
  • apply_filter=false· changes the output
  • safe=false· changes the output
Outputs
edges
CUDA kernels
  • cv_resample
Apple (MPS)
fp32, fp16
Full reference, generated from the repository

PiDiNet (Su et al., ICCV 2021): a small, fast edge detector with pixel-difference convolutions, folded into plain convolutions here; soft edges for the soft-edge ControlNet. Research use only (non-commercial).

Commercial use: no (code: LicenseRef-research-only (noncommercial); weights: LicenseRef-research-only (assumed from code)).

variantinputbackendtime (ms)peak memory (MiB, allocated / reserved)
pidinetsynthetic (1080x1920)original
portable
kernels
35.9
13.1
–
339 / 368
328 / 372
–
pidinetsynthetic (1024x1024)original
portable
kernels
portable + torch.compile
TensorRT fp16
18.5
6.59
–
5.66
4.50
199 / 236
186 / 228
–
166 / 190
187 / 200
Details
  • Original code: https://github.com/huggingface/controlnet_aux at b1aac318b15c
  • Code licence: LicenseRef-research-only (noncommercial); files: LICENSE
  • Weights licence: LicenseRef-research-only (assumed from code)
  • Licence evidence: https://github.com/huggingface/controlnet_aux/blob/b1aac318b15c4e65a21bc516dc236ad4bf593d46/src/controlnet_aux/pidi/LICENSE ; https://github.com/hellozhuo/pidinet/blob/master/LICENSE ; https://huggingface.co/lllyasviel/Annotators
  • Training data: Not stated for these checkpoints (lllyasviel/Annotators, ControlNet 1.1). The file name matches the paper's table 5 BSDS500 model.
  • Options: short_side = 512 (>= 64; also resolution; also detect_resolution), apply_filter = False, safe = False
  • Kernel libraries: cv_resample
  • Variants: pidinet
  • 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): pidinet 1080x1920 (0.3 GiB peak allocated, 0.4 reserved)
  • pidinet 1024x1024, first call: portable + torch.compile 12 s (cold Inductor cache), TensorRT fp16 20 s (ONNX + engine).
  • pidinet original 1080x1920, original 1024x1024: torch.autocast(float16) around the call; huggingface/controlnet_aux@b1aac318b15c pidinet, host OpenCV resize, float post-processing
  • pidinet load / first call: original 0.28 s / 0.058 s, portable 0.16 s / 0.014 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
pidinet0.00019 / 0.00192RTX 5060 Ti, torch 2.10.0+cu130cv_resample@969307ab22a7:sm_120

Example output

No example rendered for this checkpoint yet.