Overview
Computer vision system for classifying and localizing freight parcel damage. Built and evaluated Faster R-CNN with custom annotated dataset using Detectron2.
This project focused on designing and implementing an object detection pipeline to automate damage identification and classification in freight packages.
Key responsibilities and achievements:
- Created a custom annotated dataset of parcel boxes categorized into damage types A, B, and C
- Used Roboflow for dataset preprocessing and splitting into train/validation/test sets
- Implemented Faster R-CNN using Detectron2 with pre-trained X101-32x8d FPN architecture
- Evaluated model performance with COCO Evaluator and custom accuracy metrics
Technologies used: Detectron2, Roboflow, Pytorch, Tensorflow, OpenCV
