Adaptive Feature Pooling for Recovering Relationships between Proposal Regions and Feature Hierarchies

To recover the disrupted relationships between each proposal region and all feature hierarchy levels, we propose an adaptive feature pooling method. This method adaptively adjusts the size and position of the feature pooling to accommodate different relationships between proposal regions and feature levels, ensuring information integrity and consistency. Through this approach, we can effectively restore the connections between proposal regions and feature levels, thereby improving the accuracy and robustness of object detection.