R-CNN minus R

By Karel Lenc, Andrea Vedaldi. 2015

97 citations recorded by the source corpus.

Deep convolutional neural networks (CNNs) have had a major impact in most areas of image understanding, including object category detection. In object detection, methods such as R-CNN have obtained excellent results by integrating CNNs with region proposal generation algorithms such as selective search. In this paper, we investigate the role of proposal generation in CNN-based detectors in order to determine whether it is a necessary modelling component, carrying essential geometric information not contained in the CNN, or whether it is merely a way of accelerating detection. We do so by designing and evaluating a detector that uses a trivial region generation scheme, constant for each image. Combined with SPP, this results in an excellent and fast detector that does not require to process an image with algorithms other than the CNN itself. We also streamline and simplify the training of CNN-based detectors by integrating several learning steps in a single algorithm, as well as by proposing a number of improvements that accelerate detection.

Subfields: 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 Computer Vision and Pattern Recognition

Citations in the corpus

Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks (2015, neutral): “Similar observations are reported in [2], [39], where replacing SS region proposals with sliding windows leads to 6 percent degradation in both papers.”
Object Detection Networks on Convolutional Feature Maps (2015, methodology): “In recent detection systems [12], [13], [14], [23], [24], conv layers in the pre-trained models are thought of as region-independent feature extractors, and thus are shared on the entire image without being aware of the regions that are of interest.”
You Only Look Once: Unified, Real-Time Object Detection (2015, methodology): “R-CNN minus R replaces Selective Search with static bounding box proposals [20].”
R-FCN: Object Detection via Region-based Fully Convolutional Networks (2016, methodology): “OverFeat [22] detects objects by sliding multi-scale windows on the shared convolutional feature maps; similarly, in Fast R-CNN [7] and [13], sliding windows that replace region proposals are investigated.”
Synthetic Data for Text Localisation in Natural Images (2016, methodology): “[29] drop the region proposal stage altogether and use a constant set of regions learnt through K-means clustering on the PASCAL VOC data.”
Deep Sliding Shapes for Amodal 3D Object Detection in RGB-D Images (2015, background): “Second, 3D physical object bounding boxes vary more in size than 2D pixel-based bounding boxes (due to photography and dataset bias) [16].”
CRAFT Objects from Images (2016, methodology): “Besides these improvements built on the two-step framework, there are also some works [22, 30, 26] on end-to-end detection framework that drops the proposal step.”
Object Detection in 20 Years: A Survey (2019, background): “As “object proposal” has revolutionized the sliding window detection and has quickly dominated the deep learning based detectors, in 2014-2015, many researchers began to ask the following questions: what is the main role of the object proposals in detection? Is it for improving accuracy, or simply f”
DeepID-Net: Object Detection with Deformable Part Based Convolutional Neural Networks (2017, unknown): “Because of the power on learning feature representation, deep models have been widely used for object recognition, detection and other vision tasks [25], [52], [75], [21], [53], [85], [19], [27], [16], [35], [40], [76], [77], [33], [29], [57], [59], [57], [58], [31], [32], [30], [80], [79], [68], [6”
LocNet: Improving Localization Accuracy for Object Detection (2015, unknown): “This scheme starts from an initial set of candidate boxes B 1 (which could be, e.g ., either dense sliding windows [30, 25, 27, 22] or category-agnostic bounding box proposals [40, 35, 28]) and on each iteration t it uses the two basic components in the following way: such that those boxes they…”
Factors in Finetuning Deep Model for Object Detection with Long-Tail Distribution (2016, background): “Deep learning is applied for object detection in many works [12, 30, 18, 14, 35, 47, 43, 11, 28, 17, 27, 25, 26, 45, 15].”
Deep Learning for Generic Object Detection: A Survey (2018, methodology): “sing 300 proposals per image. The initial Faster RCNN in [229] contains several alternating training stages, later simpliﬁed in [230]. Concurrent with the development of Faster RCNN, Lenc and Vedaldi [151] challenged the role of region proposal generation methods such as selective search, studied th”
Microscopy cell counting and detection with fully convolutional regression networks (2018, unknown): “2015), bounding box regression with CNNs features (Lenc and Vedaldi 2015).”
Object Detection With Deep Learning: A Review (2018, methodology): “[88] provided an analysis on the role of proposal generation in CNN-based detectors and tried to replace this stage with a constant and trivial region generation”
G-CNN: An Iterative Grid Based Object Detector (2015, background): “There have been attempts[15, 12] to take this pre-processing stage out of the loop but they lead to performance degradations.”
Factors of Transferability for a Generic ConvNet Representation (2014, background): “Different classifiers have similar performances (also observed in [44], [45]).”