Inside-Outside Net: Detecting Objects in Context with Skip Pooling and Recurrent Neural Networks

By Sean Bell, C. Lawrence Zitnick, Kavita Bala, Ross B. Girshick. 2015

1,272 citations recorded by the source corpus.

It is well known that contextual and multi-scale representations are important for accurate visual recognition. In this paper we present the Inside-Outside Net (ION), an object detector that exploits information both inside and outside the region of interest. Contextual information outside the region of interest is integrated using spatial recurrent neural networks. Inside, we use skip pooling to extract information at multiple scales and levels of abstraction. Through extensive experiments we evaluate the design space and provide readers with an overview of what tricks of the trade are important. ION improves state-of-the-art on PASCAL VOC 2012 object detection from 73.9% to 77.9% mAP. On the new and more challenging MS COCO dataset, we improve state-of-the-art from 19.7% to 33.1% mAP. In the 2015 MS COCO Detection Challenge, our ION model won "Best Student Entry" and finished 3rd place overall. As intuition suggests, our detection results provide strong evidence that context and multi-scale representations improve small object 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

Relation Networks for Object Detection (2017, methodology): “Most ablation experiments report detection accuracies on a subset of 5k unused val images (denoted as minival) as common practice [2, 32].”
Feature Pyramid Networks for Object Detection (2016, methodology): “We train using the union of 80k train images and a 35k subset of val images ( trainval35k [2]), and report ablations on a 5k subset of val images ( minival ).”
Bridging the Gap Between Anchor-Based and Anchor-Free Detection via Adaptive Training Sample Selection (2019, methodology): “After that, lots of algorithms are proposed to improve its performance, including architecture redesign and reform [4, 9, 5, 28, 30], context and attention mechanism [2, 51, 38, 7, 44], multiscale training and testing [54, 41], training strategy and loss function [40, 52, 61, 17], feature fusion and”
Panoptic Feature Pyramid Networks (2019, background): “the box branch in Mask R-CNN benefits from the mask branch [24], and joint detection and semantic segmentation of thing classes also shows gains [3, 8, 17, 46].”
Beyond Skip Connections: Top-Down Modulation for Object Detection (2016, methodology): “For example, [4, 10, 22, 43] combine features from multiple layers for the ﬁnal clas-siﬁer; while [4, 43] use subsampled features from ﬁner scales, [10, 22] upsample the features to the ﬁnest scale and use their combination.”
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 [3, 30, 29, 40, 12, 5] (including SSD and R-FCN).”
R-FCN: Object Detection via Region-based Fully Convolutional Networks (2016, methodology): “There have been a series of orthogonal extensions of FCNs that were developed for semantic segmentation ( e.g. , see [2]), as well as extensions of region-based methods for object detection ( e.g. , see [10, 1, 23]).”
Learning to Refine Object Segments (2016, methodology): “We emphasize that DeepMask was the previous state-of-the-art on this task, outperforming all bottom-up proposal methods as well as Region Proposal Networks (RPN) [7] (we obtained improved RPN proposals from the authors of [8]).”
MegDet: A Large Mini-Batch Object Detector (2017, methodology): “We can roughly divide existing CNN-based detectors into two categories: one-stage detectors like SSD [26], YOLO [29, 30] and recent RetinaNet [24], and two-stage detectors [33, 1] like Faster RCNN [31], R-FCN [6] and Mask-RCNN [14].”
Focal Loss for Dense Object Detection (2017, methodology): “For training, we follow common practice [4], [37] and use the”
Squeeze-and-Excitation Networks (2017, background): “Further work has sought to better model spatial dependencies [7], [8] and incorporate spatial attention into the structure of the network [9].”
A MultiPath Network for Object Detection (2016, methodology): “[3] proposed to revisit skip connections for general object detection.”
Path Aggregation Network for Instance Segmentation (2018, methodology): “It is simpler compared with the process of [4, 62, 31], where L-2 normalization, concatenation and dimension reduction are needed.”
UberNet: Training a Universal Convolutional Neural Network for Low-, Mid-, and High-Level Vision Using Diverse Datasets and Limited Memory (2016, methodology): “Skip-layer normalization: Modifying slightly [5, 66], we use batch normalization [43] prior to forming the inner product with intermediate layers; this alleviates the need for very low learning rates, which was the case in [49, 107].”
Gather-Excite: Exploiting Feature Context in Convolutional Neural Networks (2018, background): “This may be one factor explaining why improving the use of context in deep networks can lead to better performance, as has been repeatedly demonstrated in object detection and other applications [1, 26, 48].”
BAM: Bottleneck Attention Module (2018, methodology): “This phenomenon can be interpreted in terms of contextual reasoning, which is widely exploited in dense prediction tasks [4, 5, 34, 47, 50].”