An Implementation of Faster RCNN with Study for Region Sampling

By Xinlei Chen, Abhinav Kumar Gupta. 2017

167 citations recorded by the source corpus.

We adapted the join-training scheme of Faster RCNN framework from Caffe to TensorFlow as a baseline implementation for object detection. Our code is made publicly available. This report documents the simplifications made to the original pipeline, with justifications from ablation analysis on both PASCAL VOC 2007 and COCO 2014. We further investigated the role of non-maximal suppression (NMS) in selecting regions-of-interest (RoIs) for region classification, and found that a biased sampling toward small regions helps performance and can achieve on-par mAP to NMS-based sampling when converged sufficiently.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging

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

In Radiology, Nuclear Medicine and Imaging

Citations in the corpus

Objects as Points (2019, neutral): “…ˆ Y x i y i c as a measure of its detection conﬁdence, and produce a bounding box at location [2] object size [2] 3D size [3] depth [1] orientation [8] joint locations where ( δ ˆ x i , δ ˆ y i ) = ˆ O ˆ x i , ˆ y i is the offset prediction and ( ˆ w i , ˆ h i ) = ˆ S ˆ x i , ˆ y i is the size…”
MAttNet: Modular Attention Network for Referring Expression Comprehension (2018, methodology): “Our implementation of Mask R-CNN6 is based on the single-GPU Faster R-CNN implementation [5].”
BAM: Bottleneck Attention Module (2018, methodology): “Deep learning has been a powerful tool for a series of pattern recognition applications including classiﬁcation, detection, segmentation and control problems.”
CBAM: Convolutional Block Attention Module (2018, methodology): “Our training code is based on [41] and we train the network for 490K iterations for fast performance validation.”
Neural Motifs: Scene Graph Parsing with Global Context (2017, methodology): “Our detector is built upon ResNet-101 [14, 4] and takes as input images of dimension 592x592, zero-padded.”
ViTPose: Simple Vision Transformer Baselines for Human Pose Estimation (2022, methodology): “* denotes using a stronger Faster RCNN [11] detector.”
RMPE: Regional Multi-person Pose Estimation (2016, methodology): “To illustrate the problems of previous approaches, we applied the state-of-the-art object detector Faster-RCNN [29] and the SPPE Stacked Hourglass model [23].”
Tracking Without Bells and Whistles (2019, background): “However, due to the changes suggested in [11] we achieve less false positives and negatives on mostly small targets.”
Lifelong Object Detection (2020, methodology): “Following the setup in (He et al. 2016; Chen and Gupta 2017), the RPN module is inserted before the last stride convolution layers of ResNet-50.”
Adversarial Examples that Fool Detectors (2017, methodology): “In contrast, a detector, like Faster RCNN [24, 5], identiﬁes bounding boxes that are “worth labelling”, and then generates labels (which might include background ) for each box.”
Spatial Memory for Context Reasoning in Object Detection (2017, methodology): “ections (N=5;10), the AP is 23:8% and 27:1%. Therefore, while we do outperform baseline for ﬁxed number of detections per image, due to limited roll-out capability we are still ˘1% below the baseline [12]. 6.1. SMN for Hard Examples In this section, we want to go beyond N=10 detections and see if th”