DeepProposal: Hunting Objects by Cascading Deep Convolutional Layers

By Amir Ghodrati, Ali Diba, Marco Pedersoli, Tinne Tuytelaars, Luc Van Gool. 2015

122 citations recorded by the source corpus.

In this paper we evaluate the quality of the activation layers of a convolutional neural network (CNN) for the generation of object proposals. We generate hypotheses in a sliding-window fashion over different activation layers and show that the final convolutional layers can find the object of interest with high recall but poor localization due to the coarseness of the feature maps. Instead, the first layers of the network can better localize the object of interest but with a reduced recall. Based on this observation we design a method for proposing object locations that is based on CNN features and that combines the best of both worlds. We build an inverse cascade that, going from the final to the initial convolutional layers of the CNN, selects the most promising object locations and refines their boxes in a coarse-to-fine manner. The method is efficient, because i) it uses the same features extracted for detection, ii) it aggregates features using integral images, and iii) it avoids a dense evaluation of the proposals due to the inverse coarse-to-fine cascade. The method is also accurate, it outperforms most of the previously proposed object proposals approaches and when plugged into a CNN-based detector produces state-of-the-art detection performance.

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

Object-Proposal Evaluation Protocol is ‘Gameable’ (2015, background): “The algorithms that fall under this category are [1, 4, 5, 10, 45, 48].”
Attend Refine Repeat: Active Box Proposal Generation via In-Out Localization (2016, background): “…scoring paradigm, a large set of image boxes is ranked according to how likely it is for each image box to tightly enclose an object — regardless of its category — and then this set is post-processed with a non-maximum-suppression step and truncated to yield the ﬁnal set of object proposals.”
Region Proposal by Guided Anchoring (2019, background): “There have been attempts [8, 9, 23, 31, 33, 34, 1, 2] that apply cascade architecture to reject easy samples at early layers or stages, and regress bounding boxes iteratively for progressive refinement.”
Advanced Deep-Learning Techniques for Salient and Category-Specific Object Detection: A Survey (2018, methodology): “Since then, several advanced deep-learningbased techniques [17]–[20] have been proposed for high-quality object detection, which covers all three related areas of OD, SOD, and COD.”
Monocular 3D Object Detection for Autonomous Driving (2016, methodology): “Parallel to our work, [18] introduced a method that generates object proposals by cascading the layers of the convolutional neural network.”
Object Detection in 20 Years: A Survey (2019, background): “and 3) neural network based approaches [123–128].”
Deep Learning for Generic Object Detection: A Survey (2018, methodology): “S15 The ﬁrst to generate object proposals by sharing full image convolutional features with detection. Most widely used object proposal method. Signiﬁcant improvements in detection speed. DeepProposal[81] VGG16 Fast RCNN 0:74 (100) 0.92 (1000) 0:58 (100) 0.80 (1000) 0:12 (100) 0.16 (1000) 53:2 (100)”
3D Object Proposals Using Stereo Imagery for Accurate Object Class Detection (2016, methodology): “[14] computes integral image features from inverse cascading layers of CNN for candidate box scoring and refinement.”
Recent Advances in Deep Learning for Object Detection (2019, background): “developed DeepProposals [91] which pre- 791 dicted proposals on the low-resolution deeper layer feature map.”
Rotation-Insensitive and Context-Augmented Object Detection in Remote Sensing Images (2018, background): “The shallow convolutional layers can better preserve ﬁne details, but they lack in semantic information [54].”
Unsupervised Learning of Discriminative Attributes and Visual Representations (2016, methodology): “The computational costs of [4, 8] respectively stem from refining proposals in a coarse-tofine inverse cascading and re-ranking EdgeBox’s proposFigure 5.”