PitchPose Studio

Simple Gradio web app for pose tracking and detection on local videos/images (YOLOv8 backend).

Project Files

• app_gradio_yolo.py — main web app
• README.md — setup and usage

Requirements

• Windows (PowerShell) or Linux
• Python 3.10+
• ultralytics, opencv-python, gradio, Pillow

Install (Windows PowerShell)

cd C:\Users\adams\Documents\Projects
python -m venv .venv-yolo
.\.venv-yolo\Scripts\Activate.ps1
python -m pip install -U pip
pip install ultralytics opencv-python gradio pillow

Run

cd C:\Users\adams\Documents\Projects
.\.venv-yolo\Scripts\Activate.ps1
python .\PoseTrackWeb\app.py

Open:

• http://127.0.0.1:7878

How To Use

1. Upload image or video.
2. Choose mode:

• predict for one-pass detection
• track for tracking IDs (ByteTrack)

3. For pose tracking, choose yolov8n-pose.pt.
4. Click Run Tracking.

Output Panels

• Animated GIF Preview (top, browser-friendly)
• Output GIF/Image Preview
• Output File (download)
• Output Video (optional) (collapsed; may fail in browser codec)
• Run Log

Output Location

Each run writes to temp folders, shown in log:

• run dir: %TEMP%\gradio_yolo_runs\run_*
• preview/download files: %TEMP%\gradio_yolo_outputs\...

Notes

• Browser video playback depends on codec. GIF preview is the most reliable web preview.
• If output does not appear, check Run Log for exact file paths.
• If app port is busy, run with another port in code (server_port=7878).

Troubleshooting

• If ultralytics import failed, verify env is active and reinstall dependencies.
• If no preview, use Output File (download) and open locally.
• If slow on laptop CPU, use yolov8n.pt and lower imgsz (e.g., 640).

GUI Screenshot

YOLO model tracking soccer video clips: