- Removed restrictive filtering - now detects 79/80 classes by default
- Added granular control: exclude_person, exclude_animals, exclude_vehicles
- Added min_confidence parameter for quality control
- Created COCO_CLASSES.md reference with all 80 detectable classes
- Better logging showing included/excluded classes and counts
- Can detect: furniture, electronics, food, accessories, sports equipment, etc.
- Default: Only excludes person (79 classes available)
- Much more flexible for diverse object detection scenarios
- Perfect for extracting multiple objects from rich scenes
- Added exclude_person parameter (default: True) to skip person detection
- Added objects_only parameter (default: True) to focus on inanimate objects
- Filters out living beings: person, animals (bird, cat, dog, etc.)
- Now extracts actual objects like tie, bag, phone, watch, etc.
- Updated node description to 'Object Extraction' instead of detection
- Changed category to 'image/object_extraction'
- Added detailed logging of filtered/excluded classes
- Perfect for extracting multiple objects without people in frame
- Added resize_with_padding() method to maintain aspect ratios
- Objects are now centered with black padding instead of stretched
- Added preserve_aspect_ratio parameter (default: True)
- Users can choose between aspect-preserving (padding) or stretching
- Prevents distortion of detected objects
- Better visual quality for cropped objects
- Moved max_size from required to optional parameters
- Added default value of 512 to prevent validation errors
- Existing workflows will now work without max_size parameter
- New workflows can optionally use max_size for size control
- Enhanced tensor shape verification with detailed logging
- Added failsafe tensor interpolation for shape mismatches
- Comprehensive error handling with fallback options
- Added reload_node.py script for development
- Created TROUBLESHOOTING.md with step-by-step fixes
- Detailed console output for debugging tensor issues
- Prevents crashes by returning single object as fallback
- Resize all cropped objects to same dimensions before batching
- Add max_size parameter to control output resolution
- Add debug logging for detection process
- Prevents 'Sizes of tensors must match' error when concatenating
- Now successfully outputs batch of detected objects
- Implements YOLOv8-based object detection with batch output
- Supports multiple YOLOv8 model sizes (nano to extra-large)
- Configurable confidence threshold, IoU threshold, and padding
- Outputs separate images for each detected object
- Includes installation scripts for Windows, Linux, and Mac
- Complete documentation and example workflow
- MIT license for open source distribution
Eduardo Ottone