Initial Release

[Xu-Justin]

Patchmentation Core Function

• Collections
  • BBox
  • Mask
  • Image
  • Patch
  • ImagePatch
  • Dataset
• Filtering
  • Filter by width
  • Filter by height
  • Filter by aspect ratio
• Transform
  • Resize
  • Random resize
  • Scale
  • Random scale
  • Grayscale
  • Random grayscale
• Visibility thresholding
• Patch distribution
• Patch blending mode
  • Hard edge
  • Soft edge
• Negative patching
• Background patching
  Description: Sometimes background already has positive patches. We want to preserve it.

Dataset

• Dataset loader
  • Loader for YOLO format dataset
  • Loader for COCO format dataset
  • Loader for Pascal VOC format dataset
• Default dataset
  • COCO 2017
  • COCO 2014
  • PASCAL VOC 2007
  • PASCAL VOC 2012
  • Penn-Fudan Database for Pedestrian Detection and Segmentation

Automation

• GitHub actions test
• PyPI deployment
  • Deployment script
  • Deployment automation with GitHub action

Additional Tasks

• README for patchmentation
• Documentation
• Docstrings

Benchmarking

• Benchmarking on Pascal VOC / COCO dataset using YOLOv?
• Custom dataset for benchmarking
• Benchmarking on a custom dataset using YOLOv?

Development Tools / Optimization (Optional)

• Visualization using Streamlit
• Conditional package installation (reference)
• Docker environment
• Implement caching
  • Shape, width, height, channel
• Fix code formatting
• Implements property / descriptors to collections
• Account gamma corrections when blending colors (reference)
• Optimize mask saving for image with no transparency
• Add progress bar when loading a dataset