Skip to content

Main Features

File and Folder Selection

  • Browse and select individual image or video files
  • Browse and select entire folders containing multiple images and videos

Model Selection

Choose from multiple models:

Face Detection

  • YOLOv8n-face (Nano): fastest inference, smallest size (~2.7 MB); lower accuracy; ideal for real-time or limited resources.
  • YOLOv8m-face (Medium): balanced speed and accuracy (~27.3 MB); solid default for most tasks.
  • YOLOv8l-face (Large): highest accuracy within v8 (~59.2 MB); slower inference; best for high precision.
  • YOLOv11m-face (Medium): newer generation with improved accuracy/speed trade-off; good general-purpose choice on modern hardware.
  • YOLOv11l-face (Large): higher accuracy variant; increased compute and memory cost.
  • YOLOv12l-face (Large): latest large model; highest accuracy and resource use; recommended for offline batch processing.
  • RetinaFace: alternative architecture with facial landmarks; good speed/accuracy for feature localization. Note: available on Apple Silicon (arm64) macOS only. Source: https://github.com/serengil/retinaface.

Hand Detection

  • HandObject (100DOH baseline) - Hand_object_detector model by ddshan, trained on 100DOH dataset
  • HandObject (100DOH TinyExplorer-tuned) - TinyExplored-Tuned version of the 100DOH hand_object_detector, targeted at detecting infant hands with ownership classification integrated

Automatic Speech Recognition

  • Whisper (OpenAI)
  • Faster Whisper
  • WhisperX

Available in sizes from tiny to large-v3-turbo. Smaller sizes are faster but have lower accuracy; larger sizes are more accurate but computationally heavier.

The performance of these models has not been assessed by the Cardiff Babylab. More information can be found in Model evaluation

Model Sources

The app automatically downloads required model weights when needed.

Sampling Rate (for face and hand detection)

  • Current sampling rate of 1 frame per second (1fps)
  • Coming soon: adjustable slider for sampling rate

Confidence Threshold Adjustment

  • Adjustable slider from 0.0 to 1.0
  • Default confidence values tailored to each model

Pipeline details

  • Face and hand detection works with images and videos
  • Batch processing for multiple files in a folder
  • Real-time progress bar and percentage display
  • Detailed logging of the recognition process

Results and Output

  • Timestamped results folder
  • CSV output with detailed detection data
  • Summary CSV with overall statistics
  • Visual results saved for images and video frames
  • Results folder opens automatically when processing completes

User Interface

  • Intuitive GUI with file/folder selection, model choice, and confidence adjustment
  • Real-time status updates in the window
  • Error handling and user notifications

Supported File Formats

Images - JPEG (.jpg, .jpeg) - PNG (.png) - BMP (.bmp)

Videos - MP4 (.mp4) - AVI (.avi) - MOV (.mov)

You can select individual files or folders containing these formats for processing.

Made with Material for MkDocs