A MultiPath Network for Object Detection

By Sergey Zagoruyko, Adam Lerer, Tsung-Yi Lin, Pedro H. O. Pinheiro, Sam Gross, Soumith Chintala, Piotr Doll'ar. 2016

218 citations recorded by the source corpus.

The recent COCO object detection dataset presents several new challenges for object detection. In particular, it contains objects at a broad range of scales, less prototypical images, and requires more precise localization. To address these challenges, we test three modifications to the standard Fast R-CNN object detector: (1) skip connections that give the detector access to features at multiple network layers, (2) a foveal structure to exploit object context at multiple object resolutions, and (3) an integral loss function and corresponding network adjustment that improve localization. The result of these modifications is that information can flow along multiple paths in our network, including through features from multiple network layers and from multiple object views. We refer to our modified classifier as a "MultiPath" network. We couple our MultiPath network with DeepMask object proposals, which are well suited for localization and small objects, and adapt our pipeline to predict segmentation masks in addition to bounding boxes. The combined system improves results over the baseline Fast R-CNN detector with Selective Search by 66% overall and by 4x on small objects. It placed second in both the COCO 2015 detection and segmentation challenges.

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

Wide Residual Networks (2016, methodology): “We also used WRN-34-2 to participate in COCO 2016 object detection challenge, using a combination of MultiPathNet [32] and LocNet [7].”
Feature Pyramid Networks for Object Detection (2016, background): “Some results were not available on the test-std set, so we also include the test-dev results (and for Multipath [40] on minival).”
Cascade R-CNN: High Quality Object Detection and Instance Segmentation (2019, methodology): “[62] proposed an ensemble of classifiers”
Fully Convolutional Instance-Aware Semantic Segmentation (2016, methodology): “Joint Mask Prediction and Classification For the instance-aware semantic segmentation task, not only [5], but also many other state-of-the-art approaches, such as SDS [15], Hypercolumn [16], CFM [7], MNC [8], and MultiPathNet [42], share a similar structure: two subnetworks are used for object segme”
Pixelwise Instance Segmentation with a Dynamically Instantiated Network (2017, background): “Another related approach [11, 49] is to use segment-based region proposals [9, 35, 36] instead of box-based proposals.”
Speed/Accuracy Trade-Offs for Modern Convolutional Object Detectors (2016, background): “Since appearing in 2015, Faster R-CNN has been particularly influential, and has led to a number of follow-up works [2, 34, 33, 45, 13, 5, 19, 44, 23, 46] (including SSD and R-FCN).”
Deep Watershed Transform for Instance Segmentation (2016, methodology): “Additionally, [31] uses a modified R-CNN model to propose instance bounding boxes, which are then further refined to obtain instance level segmentation.”
InstanceCut: From Edges to Instances with MultiCut (2016, methodology): “As an example, [16, 53] address these challenges with a complex multi-loss cascade CNN architectures, which are, however, difficult to train.”
Learning to Refine Object Segments (2016, methodology): “In the following experiments we coupled SharpMask proposals with two classifiers: VGG [26] and MultiPathNet (MPN) [41], which introduces a number of improvements to the VGG classifier.”
Cascade R-CNN: Delving Into High Quality Object Detection (2017, background): “This is closely related to the integral loss of [38], where � = {0.”
Deformable ConvNets V2: More Deformable, Better Results (2018, methodology): “In a multi-path network for object detection [39], multiple RoIpooling layers are employed for each input RoI to better exploit multi-scale and context information.”
Troubling Trends in Machine Learning Scholarship (2018, background): “t learning [30] and generative adversarial networks [51]. See [68] for more discussion of lapses in empirical rigor and resulting consequences. In contrast, many papers perform good ablation analyses [41, 45, 77, 82], and even retrospective attempts to isolate the source of gains can lead to new dis”
Weakly-Supervised Learning of Visual Relations (2017, background): “While a great progress has been made on the detection and localization of individual objects [40, 52], it is now time to move one step forward towards understanding complete scenes.”
Attend Refine Repeat: Active Box Proposal Generation via In-Out Localization (2016, methodology): “…of the other competing methods, we observe that: • Comparing with the other VGG16-Net based object detection systems (ION [3] and MultiPath [42] systems), our detection system achieves the highest COCO-style average precision with its main novelties w.r.t. the Fast R-CNN [15] baseline…”
MaskLab: Instance Segmentation by Refining Object Detection with Semantic and Direction Features (2017, background): “, [24, 81, 75], and our model is also able to learn this.”
A-Fast-RCNN: Hard Positive Generation via Adversary for Object Detection (2017, methodology): “We follow the Torch implementation [44] of Fast-RCNN.”
An Analysis of Scale Invariance in Object Detection - SNIP (2017, background): “For example, features from the layers near to the input, referred to as shallow(er) layers, are combined with deeper layers for detecting small object instances [23, 34, 1, 13, 27], dilated/deformable convolution is used to increase receptive fields for detecting large objects [32, 7, 37, 8], indepe”
Recent Advances in Deep Learning for Object Detection (2019, methodology): “Some other efforts [150, 151] focus on designing a uni-ﬁed framework with modiﬁed objective functions.”
Contextual Region-Based Convolutional Neural Network with Multilayer Fusion for SAR Ship Detection (2017, background): “In order to obtain a better performance, the IRNN which is Recurrent Neural Networks with ReLU recurrent transitions, needs to be trained on extra semantic segmentation labels, which increases the difficulty of training.”