Poseji-version2

---

Model Structure and Differences: Version 1 vs. Version 2

Version 1:

• Models Supported:
  • OpenPose: Uses OpenCV’s DNN module to load a frozen graph (graph_opt.pb) for keypoint detection.
  • MediaPipe Pose: Leverages Google’s MediaPipe Pose solution for efficient and real‑time pose detection.
  • MoveNet: Utilizes TensorFlow Lite’s MoveNet model (movenet_lightning_fp16.tflite) for fast and accurate pose estimation.
• Functionality:
  • Basic pose detection in images and videos.
  • Webcam integration and basic metrics display.
  • Session management and data export as ZIP files.

Version 2:

• Models Supported:
  Version 2 continues to support the same models as Version 1, with updates to paths and improved inference (smoothing and optional preprocessing) for more stable keypoint detection.
• New & Enhanced Features:
  • Comparison Mode: Compare two images side by side with detailed metrics and visualizations.
  • Improved UI/UX: A modern layout with custom CSS, enhanced sidebar instructions, and session management.
  • Expanded Analysis Modes: More robust modes such as advanced detailed metrics, 3D visualization, and performance optimizations for video processing.
• Code Organization:
  The code is further modularized under the poseji_v2/modules/ directory, making it easier to maintain and extend.
• Performance:
  Better smoothing of keypoints and reduced flicker in real‑time metrics; efficient media handling via temporary files and dynamic frame rate adjustments.

---