An Analysis of Scale Invariance in Object Detection - SNIP

By Bharat Singh, Larry S. Davis. 2017

831 citations recorded by the source corpus.

An analysis of different techniques for recognizing and detecting objects under extreme scale variation is presented. Scale specific and scale invariant design of detectors are compared by training them with different configurations of input data. By evaluating the performance of different network architectures for classifying small objects on ImageNet, we show that CNNs are not robust to changes in scale. Based on this analysis, we propose to train and test detectors on the same scales of an image-pyramid. Since small and large objects are difficult to recognize at smaller and larger scales respectively, we present a novel training scheme called Scale Normalization for Image Pyramids (SNIP) which selectively back-propagates the gradients of object instances of different sizes as a function of the image scale. On the COCO dataset, our single model performance is 45.7% and an ensemble of 3 networks obtains an mAP of 48.3%. We use off-the-shelf ImageNet-1000 pre-trained models and only train with bounding box supervision. Our submission won the Best Student Entry in the COCO 2017 challenge. Code will be made available at http://bit.ly/2yXVg4c.

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

Cascade R-CNN: High Quality Object Detection and Instance Segmentation (2019, methodology): “Scale invariance, an important requisite for effective object detection, has also received substantial attention in the literature [2], [35], [51].”
Objects as Points (2019, background): “This eliminates the need for multiple anchors [47].”
Swin Transformer: Hierarchical Vision Transformer using Shifted Windows (2021, background): “Unlike the word tokens that serve as the basic elements of processing in language Transformers, visual elements can vary substantially in scale, a problem that receives attention in tasks such as object detection [38, 49, 50].”
CornerNet: Detecting Objects as Paired Keypoints (2018, background): “2 - - - - - D-RFCN + SNIP (Singh and Davis, 2017) DPN-98 (Chen et al.”
Deep High-Resolution Representation Learning for Visual Recognition (2019, background): “Table 11 reports the comparison of our network to state-of-the-art single-model object detectors on COCO test-dev without using multi-scale training and multi-scale testing that are done in [85], [97], [110], [115], [127], [128].”
Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation (2019, background): “Additionally, incorporating scale-aware information to feature pyramid [45] or image pyramid [65] may improve the performance of bottomup methods.”
Dynamic Head: Unifying Object Detection Heads with Attentions (2021, background): “Early works have demonstrated the significance of leveraging image pyramid methods [6, 22, 23] for multi-scale training.”
CenterNet: Keypoint Triplets for Object Detection (2019, neutral): “%) and that of CenterNet511-104 is comparable to D-RFCN + SNIP [39] (44.9% vs. 45.7%).”
Scale-Aware Trident Networks for Object Detection (2019, methodology): “Similar to SNIP [38], we define a valid range [li, ui] for each branch i.”
Libra R-CNN: Towards Balanced Learning for Object Detection (2019, methodology): “Other methods focus on cascade procedures [24], duplicate removal [14, 13], multi-scales [2, 1, 31, 30], adversarial learning [37] and more contextual [36].”
Feature Selective Anchor-Free Module for Single-Shot Object Detection (2019, neutral): “3 Deformable R-FCN + SNIP [30] DPN-98 45.”
FoveaBox: Beyound Anchor-Based Object Detection (2019, methodology): “at each pre-deﬁned sliding window anchor with a light-weight network. Several attempts have been performed to boost the performance of the detector, including feature pyramid [27][24][21], multiscale [44][34], and object relations [17], etc. Compared to two-stage approaches, the one-stage pipeline s”
Deep Regionlets for Object Detection (2017, methodology): “h as RetinaNet and Mask R-CNN, even without additional segmentation labels. 1. Introduction Generic object detection has been extensively studied by the computer vision community over several decades [22, 4, 48, 16, 17, 40, 8, 51, 29, 47, 45, 10, 13, 6, 44, 52] due to its appeal to both academic res”
YOLO-FaceV2: A Scale and Occlusion Aware Face Detector (2022, background): “Besides, SNIP [43] and TridentNet [12] also provide new ideas to solve the multi-scale problem, which will be discussed in detail in the following sections.”
Improved YOLOv5 network for real-time multi-scale traffic sign detection (2021, background): “And learning scale-invariant representation is critical for target recognition and location[20].”
RepPoints: Point Set Representation for Object Detection (2019, background): “The bestperforming object detectors to date generally follow a multistage recognition paradigm [24, 4, 13, 23, 2, 35, 27], and the bounding box representation appears in almost all stages: 1) as pre-defined [33, 25] or learnt [40, 44, 46] anchors that serve as hypotheses over the bounding box space;”