ruisv/bcdl-pidnet

PIDNet-S for RDK S100/S100P — real-time semantic segmentation

Compiled BPU models (.hbm) for the D-Robotics RDK S100 / S100P, ready to load — no ONNX export, no calibration, no hb_compile. Built and measured with BCDL, a C++17 inference and media library for the RDK S-series with Python bindings.

Upstream: PIDNet-S, Cityscapes 19 classes

Non-commercial. These weights carry a restriction the upstream code licence never mentions — see Licence at the bottom. Redistribution is permitted; commercial use is not. If you need a commercial build, the recipe is public and you can retrain or substitute the backbone.

Files

file	what it is
pidnet_s_nashm_1024x2048_nv12_v3.hbm	2048x1024 NV12, 19 classes — 18.0 MB

Measured on an S100P

stage	latency	throughput
segmentation	4.48 ms	223 FPS

hrt_model_exec perf, one thread, minimum of three runs, on a board first gated against its own previously recorded numbers. BPU time only — CPU pre/post-processing is on top and is listed per task in BCDL's benchmark results.

Use it

conda install -c https://mirrors.ruis.ai/conda -c conda-forge bcdl

import bcdl
engine = bcdl.Engine("pidnet_s_nashm_1024x2048_nv12_v3.hbm")
print(engine.input_shape(0), engine.output_shape(0))

Each task has a decoder in BCDL that turns those raw outputs into boxes, keypoints, masks, disparity or text — see the Python API (中文).

What to know before deploying

11x faster than the DeepLabV3+ build it replaces (4.48 ms against 49.6 ms) at half the model size, for a 0.9859 output cosine and 94.6% pixel agreement.

The output is at 1/8 resolution — [1, 19, 128, 256] — and the label map is upsampled by the caller. Argmax costs 0.18 ms here, so folding it into the graph would buy nothing.

The _v3 suffix is load-bearing. Earlier builds were calibrated on data that had not been pre-normalised. When cal_data_type is float32 the compiler's norm_type does not apply to the calibration data, so the input thresholds come out wrong — and the model still compiles without a warning and segments to noise. If you rebuild this, check the input threshold in quant_info.json.

Non-commercial. See the licence note above.

Licence

MIT on the code, but the weights are trained on Cityscapes, whose terms permit distributing a trained model and bar commercial use of it in the same breath.

BCDL itself is Apache-2.0 and is unrelated to these terms — it is a general-purpose runtime that loads any .hbm. The licence above constrains these weights and this compiled artefact.

The conversion recipe — ONNX export, calibration, hb_compile config and the acceptance numbers — is public in bcdl-model-zoo, so this build can be reproduced or retargeted rather than taken on trust.