ruisv/bcdl-xfeat

XFeat for RDK S100/S100P — sparse local features

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: XFeat / accelerated_features

Redistributable, including commercially. The licence chain was checked on the code, the pretrained weights it started from, and the data it was trained on — all three, because a permissive repository badge does not by itself say anything about the weights. See Licence.

Files

file	what it is
xfeat_nashm_640x480.hbm	backbone, 640x480, 3 outputs — 3.0 MB

Measured on an S100P

stage	latency	throughput
backbone	0.99 ms	1013 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("xfeat_nashm_640x480.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

Only the convolutional backbone is compiled — 3 MB of it. Keypoint NMS, top-k selection and sparse descriptor sampling stay on the CPU, which is what keeps the graph free of dynamic control flow. BCDL does that half for you.

Two rewrites were needed to export it at all, and both were checked numerically before being trusted: the input InstanceNorm was lifted out of the graph into CPU preprocessing, and _unfold2d became pixel_unshuffle (a single SpaceToDepth), asserted equal to the original beforehand.

The descriptor sampler is bicubic, not bilinear. Upstream takes the default mode of InterpolateSparse2d, which is bicubic, while the reliability map in the same file uses bilinear. Getting that wrong leaves shapes, counts and keypoints all correct and the descriptor cosine stuck at 0.9965 — which reads like quantisation noise. It is not.

Licence

Apache-2.0, and the pretrained weights are committed inside that repository — so the same grant covers them.

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.