IoU-uniform R-CNN: Breaking through the limitations of RPN
作者：
Highlights：
• We reveals the importance of solving the limitations of RPN and our proposed IoU-uniform R-CNN can alleviate the IoU distribution imbalance and inadequate training samples by generating samples with uniform IoU distribution.
• We improve the performance of IoU prediction branch by eliminating the feature offsets of RoIs at inference.
• Our proposed method consistently obtains significant improvements over multiple state-of-the-art detectors. Specially, it achieves 2.4 AP improvement than Faster R-CNN (with ResNet 101-FPN backbone) on MS COCO dataset.
摘要
•We reveals the importance of solving the limitations of RPN and our proposed IoU-uniform R-CNN can alleviate the IoU distribution imbalance and inadequate training samples by generating samples with uniform IoU distribution.•We improve the performance of IoU prediction branch by eliminating the feature offsets of RoIs at inference.•Our proposed method consistently obtains significant improvements over multiple state-of-the-art detectors. Specially, it achieves 2.4 AP improvement than Faster R-CNN (with ResNet 101-FPN backbone) on MS COCO dataset.
论文关键词：Object detection,Two-stage detector,RPN,IoU distribution imbalance
论文评审过程：Received 12 December 2019, Revised 28 December 2020, Accepted 2 January 2021, Available online 7 January 2021, Version of Record 14 January 2021.
论文官网地址：https://doi.org/10.1016/j.patcog.2021.107816
