Main points:
A project introduction
Balloon detection project based on MASK-RCNN framework
- Project Introduction
- Project Flow
- download dataset
- Download the source code of the MASK-RCNN framework
- Introduction to MASK-RCNN model
- Interpretation of FPN layer feature extraction principle
- Interpretation of FPN network architecture implementation
- Build box scale settings
- Generate all boxes based on feature maps of different scales
- Interpretation of the role and implementation of the RPN layer
- Candidate Box Filtering Method
- Proposal layer implementation method
- The role of the DetectionTarget layer
- Positive and negative sample selection and label definition
- The role and purpose of the RoiPooling layer
- Effect of the RorAlign operation
- Overall Framework Review
- Lableme tool installation
- Use Labelme to label data and labels
- MASK-RCNN source code modification method
- Training required tasks based on labeled data
- Test and Demonstration Module