CV / ViTPose

ViTPose-Small-BN

ViTPose-Small-BN is the BatchNorm variant of ViTPose-Small. Same I/O (256x192 RGB, 64x48x17 heatmap, 17 COCO keypoints) with slightly lower accuracy (AP 72.01 vs 74.16).

6 downloads

Size

31 MB

Memory

4GB+

Precision

Hailo HEF / HailoRT 4.23.x

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 \
  --name cm5-hailo8-vitpose-small-bn \
  --privileged \
  --net=host \
  -e PYTHONUNBUFFERED=1 \
  --device /dev/hailo0:/dev/hailo0 \
  -v /usr/lib/libhailort.so.4.23.0:/usr/lib/libhailort.so.4.23.0:ro \
  -v /usr/lib/libhailort.so:/usr/lib/libhailort.so:ro \
  ghcr.io/seeed-projects/recomputer-hailo8-cv/vit_pose_small_bn:latest \
  python web_detection.py --model_path model/vit_pose_small_bn.hef --video_path video/test.mp4

REST API

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

Curl

curl -X POST "http://<Board_IP>:8000/api/models/vit_pose_small_bn/predict" \
  -F "file=@test.jpg"

Python

import requests
response = requests.post(
    "http://<Board_IP>:8000/api/models/vit_pose_small_bn/predict",
    files={"file": open("test.jpg", "rb")},
    timeout=30,
)
print(response.json())

Model Details

ViTPose-Small-BN on reComputer R Series (CM5 + Hailo-8)

ViTPose-Small-BN is the BatchNorm variant of ViTPose-Small. Same I/O (256x192 RGB, 64x48x17 heatmap, 17 COCO keypoints) with slightly lower accuracy (AP 72.01 vs 74.16).

Model Info

Property	Value
Architecture	ViT-Small-BN (ViTPose)
Task	Single-person pose estimation
Input	256x192x3 RGB
Output	Heatmap 64x48x17
Parameters	24.32M
Operations	17.17G
HEF	Hailo Model Zoo v2.19.0, Hailo-8

Run With Demo Video

bash

sudo docker run --rm \
  --name cm5-hailo8-vitpose-small-bn \
  --privileged \
  --net=host \
  -e PYTHONUNBUFFERED=1 \
  --device /dev/hailo0:/dev/hailo0 \
  -v /usr/lib/libhailort.so.4.23.0:/usr/lib/libhailort.so.4.23.0:ro \
  -v /usr/lib/libhailort.so:/usr/lib/libhailort.so:ro \
  ghcr.io/seeed-projects/recomputer-hailo8-cv/vit_pose_small_bn:latest \
  python web_detection.py --model_path model/vit_pose_small_bn.hef --video_path video/test.mp4

REST API

bash

curl -X POST "http://<Board_IP>:8000/api/models/vit_pose_small_bn/predict" \
  -F "file=@test.jpg"

Endpoint	Method	Purpose
`/api/models/vit_pose_small_bn/predict`	POST	17 keypoints (JSON)
`/api/video_feed`	GET	MJPEG preview with skeleton

Development Notes

Source: src/rpi5_hailo8_vit_pose_small_bn/
Container: ghcr.io/seeed-projects/recomputer-hailo8-cv/vit_pose_small_bn:latest
Family: vit_pose (vit_pose_small / vit_pose_small_bn)

Inputs and Outputs

Input: image, video, or USB camera frame. Output: 17 COCO keypoints (x, y, score) with skeleton overlay.