# Revisiting RCNN: On Awakening the Classification Power of Faster RCNN

arXiv: https://arxiv.org/abs/1803.06799 · HTML: https://mlsanity.com/paper/1803.06799/

## Key facts

- Cited by 28 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 19 Mar 2018.

## Built on by (most cited first, 28 of 28)

- [Learning Spatio-Temporal Transformer for Visual Tracking](https://mlsanity.com/paper/2103.17154/) (2021)
- [DeFRCN: Decoupled Faster R-CNN for Few-Shot Object Detection](https://mlsanity.com/paper/2108.09017/) (2021)
- [Occluded Human Mesh Recovery](https://mlsanity.com/paper/2203.13349/) (2022)
- [FCPose: Fully Convolutional Multi-Person Pose Estimation with Dynamic Instance-Aware Convolutions](https://mlsanity.com/paper/2105.14185/) (2021)
- [End-to-end Prostate Cancer Detection in bpMRI via 3D CNNs: Effects of Attention Mechanisms, Clinical Priori and Decoupled False Positive Reduction](https://mlsanity.com/paper/2101.03244/) (2021)
- [Multi-Instance Pose Networks: Rethinking Top-Down Pose Estimation](https://mlsanity.com/paper/2101.11223/) (2021)
- [Multi-Class Cell Detection Using Spatial Context Representation](https://mlsanity.com/paper/2110.04886/) (2021)
- [Mutual Supervision for Dense Object Detection](https://mlsanity.com/paper/2109.05986/) (2021)
- [Lightweight Super-Resolution Head for Human Pose Estimation](https://mlsanity.com/paper/2307.16765/) (2023)
- [Prototype-based Cross-Modal Object Tracking](https://mlsanity.com/paper/2312.14471/) (2023)
- [BoxMask: Revisiting Bounding Box Supervision for Video Object Detection](https://mlsanity.com/paper/2210.06008/) (2022)
- [Industrial Anomaly Detection and Localization Using Weakly-Supervised Residual Transformers](https://mlsanity.com/paper/2306.03492/) (2023)
- [Recursively Refined R-CNN: Instance Segmentation with Self-RoI Rebalancing](https://mlsanity.com/paper/2104.01329/) (2021)
- [Efficient Decoder-free Object Detection with Transformers](https://mlsanity.com/paper/2206.06829/) (2022)
- [PLoc: A New Evaluation Criterion Based on Physical Location for Autonomous Driving Datasets](https://mlsanity.com/paper/2403.19893/) (2024)
- [Pseudo-IoU: Improving Label Assignment in Anchor-Free Object Detection](https://mlsanity.com/paper/2104.14082/) (2021)
- [A Survey on Deep Domain Adaptation and Tiny Object Detection Challenges, Techniques and Datasets](https://mlsanity.com/paper/2107.07927/) (2021)
- [Decoupled IoU Regression for Object Detection](https://mlsanity.com/paper/2202.00866/) (2022)
- [Enhancing Your Trained DETRs with Box Refinement](https://mlsanity.com/paper/2307.11828/) (2023)
- [E^2VTS: Energy-Efficient Video Text Spotting from Unmanned Aerial Vehicles](https://mlsanity.com/paper/2206.02281/) (2022)
- [A Global to Local Double Embedding Method for Multi-person Pose Estimation](https://mlsanity.com/paper/2102.07318/) (2021)
- [Pyramid Correlation based Deep Hough Voting for Visual Object Tracking](https://mlsanity.com/paper/2110.07994/) (2021)
- [Towards Precise Weakly Supervised Object Detection via Interactive Contrastive Learning of Context Information](https://mlsanity.com/paper/2304.14114/) (2023)
- [EXOT: Exit-aware Object Tracker for Safe Robotic Manipulation of Moving Object](https://mlsanity.com/paper/2306.05262/) (2023)
- [Metric-aligned Sample Selection and Critical Feature Sampling for Oriented Object Detection](https://mlsanity.com/paper/2306.16718/) (2023)
- [Anchor-Intermediate Detector: Decoupling and Coupling Bounding Boxes for Accurate Object Detection](https://mlsanity.com/paper/2310.05666/) (2023)
- [Self-Balanced R-CNN for Instance Segmentation](https://mlsanity.com/paper/2404.16633/) (2024)
- [Beyond Fixed Thresholds and Domain-Specific Benchmarks for Explainable Multi-Task Classification in Autonomous Vehicles](https://mlsanity.com/paper/2605.04299/) (2026)

## Abstract (arXiv)

Recent region-based object detectors are usually built with separate classification and localization branches on top of shared feature extraction networks. In this paper, we analyze failure cases of state-of-the-art detectors and observe that most hard false positives result from classification instead of localization. We conjecture that: (1) Shared feature representation is not optimal due to the mismatched goals of feature learning for classification and localization; (2) multi-task learning helps, yet optimization of the multi-task loss may result in sub-optimal for individual tasks; (3) large receptive field for different scales leads to redundant context information for small objects.We demonstrate the potential of detector classification power by a simple, effective, and widely-applicable Decoupled Classification Refinement (DCR) network. DCR samples hard false positives from the base classifier in Faster RCNN and trains a RCNN-styled strong classifier. Experiments show new state-of-the-art results on PASCAL VOC and COCO without any bells and whistles.