Subcategory-Aware Convolutional Neural Networks for Object Proposals and Detection

By Yu Xiang, Wongun Choi, Yuanqing Lin, Silvio Savarese. 2016

289 citations recorded by the source corpus.

In Convolutional Neural Network (CNN)-based object detection methods, region proposal becomes a bottleneck when objects exhibit significant scale variation, occlusion or truncation. In addition, these methods mainly focus on 2D object detection and cannot estimate detailed properties of objects. In this paper, we propose subcategory-aware CNNs for object detection. We introduce a novel region proposal network that uses subcategory information to guide the proposal generating process, and a new detection network for joint detection and subcategory classification. By using subcategories related to object pose, we achieve state of-the-art performance on both detection and pose estimation on commonly used benchmarks.

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

Objects as Points (2019, methodology): “KITTI contains 7841 training images and we follow standard training and validation splits in literature [10, 54].”
CornerNet: Detecting Objects as Paired Keypoints (2018, background): “Other works focus on incorporating sub-category information (Xiang et al., 2016), generating object proposals at multiple scales with more contextual information (Bell et al., 2016; Cai et al., 2016; Shrivastava et al., 2016; Lin et al., 2016), selecting better features (Zhai et al., 2017),…”
AugGAN: Cross Domain Adaptation with GAN-Based Data Augmentation (2018, methodology): “With the previously trained AugGAN, be it trained with SYNTHIA or GTA dataset, we transformed the KITTI dataset (7481 images with 6686 of which contains vehicle instances) [18] to its nighttime version and evaluate the translation results via detector training.”
Multi-level Fusion Based 3D Object Detection from Monocular Images (2018, methodology): “A region proposal network (RPN) is utilized to generate 2D proposals in the image, as RPN provides strong objectness con-ﬁdence regions of interest (RoIs) with CNN features and it can share weights with the down-stream detection network [31, 36, 37, 19].”
Disentangling Monocular 3D Object Detection (2019, result): “Please note that the works in [40, 24, 41] are using yet another training/validation split, rendering their results incomparable to ours while yielding numerically comparable ranges to e.”
Accurate Single Stage Detector Using Recurrent Rolling Convolution (2017, methodology): “Subcategory information was used in [21] to enhance the region propose stage and achieved promising results in KITTI.”
MonoPair: Monocular 3D Object Detection Using Pairwise Spatial Relationships (2020, methodology): “SubCNN [40] learns viewpoint-dependent subcategories from 3D CAD models to capture both shape, viewpoint and occlusion patterns.”
M3D-RPN: Monocular 3D Region Proposal Network for Object Detection (2019, methodology): “We conduct experiments on three common data splits including val1 [5], val2 [35], and the official test split [12].”
Pseudo-LiDAR++: Accurate Depth for 3D Object Detection in Autonomous Driving (2019, background): “…frontal-view detection pipeline (Ren et al., 2015; He et al., 2017; Lin et al., 2017) — fell far behind in localizing objects in 3D (Li et al., 2019a; Xiang et al., 2015; 2017; Chabot et al., 2017; Mousavian et al., 2017; Chen et al., 2015; Xu & Chen, 2018; Chen et al., 2016; Pham & Jeon, 2017;…”
DepthCN: Vehicle detection using 3D-LIDAR and ConvNet (2017, background): “[18] introduced a ConvNet-based region proposal network that uses subcategory information to guide the proposal generating process.”
3D Bounding Box Estimation Using Deep Learning and Geometry (2016, methodology): “7 increases approximately linearly with distance, compared to SubCNN’s super-linear falloff.”
SqueezeDet: Unified, Small, Low Power Fully Convolutional Neural Networks for Real-Time Object Detection for Autonomous Driving (2016, methodology): “Its mean average precision of all 3 difﬁculty levels in 3 categories outperforms the top published methods [26, 3].”
Knowledge-based Recurrent Attentive Neural Network for Traffic Sign Detection (2018, background): “One typical branch is to design multi-scale neural networks to extract features of different levels to adapt the detection of objects of diverse sizes [6], [7].”
Perception, Planning, Control, and Coordination for Autonomous Vehicles (2017, background): “In [131], by further dividing the object category into sub-categories based on their 2D appearance, 3D pose and 3D shape, the authors were able to train the deep learning network to recover both 2D and 3D information from the 2D image.”
Object Detection With Deep Learning: A Review (2018, background): “[105], Faster R-CNN [18], HyperNet [101], ION [95], MSGR [104], StuffNet [100], SSD300 [71], SSD512 [71], OHEM [113], SDP+CRC [33], GCNN [70], SubCNN [60], GBD-Net [109], PVANET [116], YOLO [17], YOLOv2 [72], R-FCN [65], FPN [66], Mask R-CNN [67], DSSD [73] and DSOD [74].”
A Survey on 3D Object Detection Methods for Autonomous Driving Applications (2019, methodology): “Despite the previous 3DVP representations [41], [45] allow to model occlusion and parts appearance, they are obtained as a classification among an existing dictionary of visibility patterns common in the training set.”
CubeSLAM: Monocular 3-D Object SLAM (2018, methodology): “The commonly used training and validation index split by [2, 13] is used.”