alt.hn

6/14/2026 at 2:37:51 PM

Show HN: Dual YOLOv8n UAV Detection on RK3588S at 42 FPS Using NPU

 https://github.com/alebal123bal/khadas_yolov8n_multithread

by alebal123bal

6/14/2026 at 4:15:26 PM

Is there something special about yolov8 over later models (9-12)? It seems most of the research and working examples default to v8 despite it being 3 years old. Or just because it is what fits on this hardware?

by robinduckett

6/14/2026 at 5:02:21 PM

Newer versions aren't open source, or at least have murky licencing.

by snovv_crash

6/14/2026 at 5:13:08 PM

Ahh that’ll do it. A shame really, the later models seem to be fairly good just from my idle testing as an enthusiast.

by robinduckett

6/14/2026 at 5:08:12 PM

Mainly because YOLOv8 is well-supported by the Rockchip/RKNN toolchain.

The goal here was an end-to-end RK3588S pipeline rather than comparing detector families: training/export, ONNX graph fixing, INT8 RKNN conversion, C++ postprocessing, and runtime inference across the 3 NPU cores. YOLOv8 has known-good export paths and Rockchip examples, so it was the most practical baseline.

Newer YOLO versions may be possible, but usually require more work around RKNN export compatibility.

by alebal123bal

6/14/2026 at 5:55:32 PM

More slop again. The way to get more throughput is to bump batch size, not to try and "multithread" job submits to the NPU as if its a CPU.

by stefan_

6/14/2026 at 2:38:19 PM

I built this while trying to understand how much of the RK3588S vision pipeline could be kept off the CPU.

The main trick is not the YOLO model itself, but the pipeline structure: MIPI capture through the ISP, resize/color conversion through RGA, and YOLOv8n inference through all 3 NPU cores with one RKNN context per core. With a 3-thread inference pool the pipeline goes from ~31 FPS to the OS08A10 camera’s 46 FPS ceiling.

The memory footprint is also small: roughly 137–152 MB RSS for one 1080p stream, using a fixed preallocated buffer pool rather than per-frame allocations. Two streams are roughly 276–304 MB RSS.

The repo also has a multi-process side of the pipeline: detections are published over Unix-domain sockets to tracking, temporal features, a presence FSM, and an optional Qwen2.5-0.5B summary step. For the LLM step, the camera pipeline can temporarily blackout/resume so RKLLM gets the whole NPU.

I split the work into three repos:

- runtime dual-stream YOLOv8n RK3588S pipeline: https://github.com/alebal123bal/khadas_yolov8n_multithread

- train/export/INT8 RKNN conversion for YOLOv8/YOLOv5: https://github.com/alebal123bal/RKNN_TRAIN_YOLO

- Qwen on RK3588S, via RKLLM/NPU or llama.cpp/CPU: https://github.com/alebal123bal/RKLLM_LLAMA_QWEN

The demo class is UAV/drone, but this is meant as a general edge-inference pipeline example, not an operational/surveillance/defense system.

by alebal123bal

6/14/2026 at 4:00:52 PM

[dead]

by ancientmoth