Revisiting RCNN: On Awakening the Classification Power of Faster RCNN

By Bowen Cheng, Yunchao Wei, Humphrey Shi, Rogério Schmidt Feris, Jinjun Xiong, Thomas S. Huang. 2018

241 citations recorded by the source corpus.

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.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Deformable ConvNets V2: More Deformable, Better Results (2018, background): “In [5], the authors find redundant context to be a plausible source of detection error for Faster R-CNN.”
Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation (2019, methodology): “Motivated by [14, 13, 54], we employ separate ASPP and decoder modules for semantic segmentation and instance segmentation, respectively, based on the hypothesis that those two branches requires different contextual and decoding information, which is empirically verified in the following section.”
Mask Scoring R-CNN (2019, background): “In [5], Cheng et al . discussed the false positive samples and used a separated network for correcting the score of such samples.”
Scale-Aware Trident Networks for Object Detection (2019, neutral): “DCR [9] ResNet-101-FPN-Deformable 43.”
Self-produced Guidance for Weakly-supervised Object Localization (2018, methodology): “Convolutional neural network has been widely used in object detection and localization tasks [16,22,23,24,25,26].”
ScratchDet: Exploring to Train Single-Shot Object Detectors from Scratch (2018, background): “Recent developments of two-stage approach focus on redesigning architecture diagram [20], convolution form [5], re-ranking detection scores [3], using contextual reasoning [1] and exploiting multiple layers for prediction [19].”
Multi-task Generative Adversarial Network for Detecting Small Objects in the Wild (2020, methodology): “Speciﬁcally, as done in (Cheng et al. 2018b), region scores (i.e. classiﬁcation score) of our MTGAN framework is aggregated with region scores of our baseline detector by element-wise product to form the ﬁnal classiﬁcation score of each region.”
Recent Advances in Deep Learning for Object Detection (2019, methodology): “In object 1380 detection, some works [132,164] also investigate this training 1381”
DeFRCN: Decoupled Faster R-CNN for Few-Shot Object Detection (2021, background): “2 may lead to possible suboptimal solution on individual tasks in order to balance the mismatched goals of them [7, 49].”
Revisiting the Sibling Head in Object Detector (2020, unknown): “2 DCR [4] ResNet-101 40.”
HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose Estimation (2019, background): “The person bounding boxes are usually generated by an object detector [36, 26, 14, 13].”
Learning From Noisy Anchors for One-Stage Object Detection (2019, background): “In addition, various hard [31, 26, 5, 24] and soft [22, 17, 3] sampling strategies have been proposed.”
Deep Regionlets for Object Detection (2017, background): “The two-stage detectors consistently achieve higher accuracy than single-stage detectors and numerous extensions have been proposed [9, 35, 18, 6, 44, 21, 7].”
Diverse Knowledge Distillation for End-to-End Person Search (2020, methodology): “It is a common practice to improve the main network using the additional guidance from external networks (Li, Jin, and Yan 2017; Cheng et al. 2018; Zhu et al. 2019; Hinton, Vinyals, and Dean 2015; Tung and Mori 2019; Liu et al. 2019; Yuan et al. 2020).”
Dynamic R-CNN: Towards High Quality Object Detection via Dynamic Training (2020, background): “Recent researches focus on improving object classiﬁer from various perspectives [26,17,31,39,22,46,6,15].”
D2Det: Towards High Quality Object Detection and Instance Segmentation (2020, unknown): “into single-stage [35, 42, 40, 29, 27, 4] and two-stage methods [18, 43, 17, 41, 8, 21].”