CV / PPSeg

PP-LiteSeg

PP-LiteSeg Cityscapes semantic segmentation accelerated by RKNN.

4 downloads

Precision

RKNN

Choose the device you're using, the set up guide and documentation will update accordingly.

Platform

Inference Engine

Getting Started

Deploy

sudo docker run --rm --privileged --net=host \
  -e PYTHONUNBUFFERED=1 -e RKNN_LOG_LEVEL=0 \
  --device /dev/dri/renderD129:/dev/dri/renderD129 \
  -v /proc/device-tree/compatible:/proc/device-tree/compatible:ro \
  ghcr.io/seeed-projects/recomputer-rk-cv/rk3576-ppseg:latest \
  python web_service.py --platform rk3576 --model_dir /app/model --camera_id -1 --host 0.0.0.0 --port 8000

REST API

Use the REST API to run inference. Copy the commands below.

Curl

curl http://localhost:8080/v1/chat/completions -d '{
  "model": "pp-liteseg-rknn",
  "messages": [{"role": "user", "content": "Hello"}]
}'

Python

import requests

resp = requests.post(
    "http://localhost:8080/v1/chat/completions",
    json={"model": "pp-liteseg-rknn", "messages": [{"role": "user", "content": "Hello"}]},
)
print(resp.json())

Model Details

Quick Start

1. Install Docker

Run the following commands on the development board to install Docker:

bash

# Download installation script
curl -fsSL https://get.docker.com -o get-docker.sh
# Install using Aliyun mirror source
sudo sh get-docker.sh --mirror Aliyun
# Start Docker and enable auto-start on boot
sudo systemctl enable docker
sudo systemctl start docker

2. Run the Project (One command, dual-mode preview)

This service packages the PP-LiteSeg Cityscapes example from reComputer-RK-CV with RKNN inference, Web preview, REST APIs, and asynchronous MP4 processing.

Model Information

Property	Value
Model	`model/ppseg.rknn`
Input	`512 x 512` RGB
Output	19-class Cityscapes segmentation mask
Warm-up image	`model/test.png`

Step A: Pull Images

bash

sudo docker pull ghcr.io/seeed-projects/recomputer-rk-cv/rk3588-ppseg:latest
sudo docker pull ghcr.io/seeed-projects/recomputer-rk-cv/rk3576-ppseg:latest

Step B: Run with One Click

For RK3576:

bash

sudo docker run --rm --privileged --net=host \
  -e PYTHONUNBUFFERED=1 -e RKNN_LOG_LEVEL=0 \
  --device /dev/dri/renderD129:/dev/dri/renderD129 \
  -v /proc/device-tree/compatible:/proc/device-tree/compatible:ro \
  ghcr.io/seeed-projects/recomputer-rk-cv/rk3576-ppseg:latest \
  python web_service.py --platform rk3576 --model_dir /app/model \
  --camera_id -1 --host 0.0.0.0 --port 8000

For RK3588:

bash

sudo docker run --rm --privileged --net=host \
  -e PYTHONUNBUFFERED=1 -e RKNN_LOG_LEVEL=0 \
  --device /dev/dri/renderD129:/dev/dri/renderD129 \
  -v /proc/device-tree/compatible:/proc/device-tree/compatible:ro \
  ghcr.io/seeed-projects/recomputer-rk-cv/rk3588-ppseg:latest \
  python web_service.py --platform rk3588 --model_dir /app/model \
  --camera_id -1 --host 0.0.0.0 --port 8000

Open http://<BOARD_IP>:8000 or /docs. Use --camera_id N with a mapped /dev/videoN node for a camera. Use --video video/test.mp4 for the bundled video; --video and --video_path are aliases and override the camera.

🔌 API Documentation

1. Segmentation Interface (Predict)

Endpoint: POST /api/models/ppseg/predict

Request Parameters (Multipart/Form-Data):

file: Image file to segment. The service resizes it to 512 x 512 before inference.

Usage Examples:

bash

curl -X POST "http://<BOARD_IP>:8000/api/models/ppseg/predict" \
  -F "file=@test.png"

The response contains inference time, source dimensions, and per-class pixel counts. The latest mask overlay is served by GET /api/video_feed.

Endpoint	Purpose
`GET /api/health`	Platform, model names, and readiness.
`GET /api/video_feed`	Latest color-mask overlay as MJPEG.
`POST /api/video/upload`	Upload one MP4 in field `file`.
`POST /api/video/analyze`	Analyze an uploaded `filename`.
`GET /api/video/status`	Background progress and errors.
`GET /api/video/list`	Uploaded and generated videos.
`GET /api/video/download/{filename}`	Download a result.

2. System Configuration Interface (Config)

The generic threshold and topk configuration fields are retained for service compatibility but do not change the fixed argmax segmentation. A replacement model must retain the expected 19-class output or update the palette and post-processing code.

The runtime resizes input to 512 x 512, converts BGR to RGB, accepts NCHW or NHWC 19-class output, restores the mask with nearest-neighbor interpolation, and overlays the Cityscapes palette. Conversion sources are under rknn_model_zoo/examples/ppseg in the upstream project.

3. Command Line Arguments

Argument	Default	Description
`--platform`	Required	`rk3576` or `rk3588`.
`--model_dir`	`model`	Directory containing `ppseg.rknn` and `test.png`.
`--camera_id`	`-1`	Camera index; `-1` disables capture.
`--video`, `--video_path`	None	Looping local video; overrides the camera.
`--host` / `--port`	`0.0.0.0` / `8000`	FastAPI listen address and port.

PYTHONUNBUFFERED=1 flushes logs immediately. RKNN_LOG_LEVEL=0 suppresses known harmless static-model initialization messages; remove it while debugging RKNN startup.

Real-time Video Stream Interface (Video Feed)

Get the latest annotated MJPEG stream for browser preview:

Endpoint: GET /api/video_feed
Example Usage: <img src="http://<BOARD_IP>:8000/api/video_feed">

🛠️ Developer Guide (Production Recommendations)

Build Local Images

bash

docker build -f docker/rk3576/ppseg.dockerfile \
  -t rk3576-ppseg:local src/rk3576_ppseg

docker build -f docker/rk3588/ppseg.dockerfile \
  -t rk3588-ppseg:local src/rk3588_ppseg

Inputs and Outputs

Input: image, video frame, or camera frame. Output: Cityscapes mask, class pixel counts, and overlay.