Scale-Transferrable Object Detection

By Peng Zhou, Bingbing Ni, Cong Geng, Jianguo Hu, Yi Xu. 2018

327 citations recorded by the source corpus.

Scale problem lies in the heart of object detection. In this work, we develop a novel Scale-Transferrable Detection Network (STDN) for detecting multi-scale objects in images. In contrast to previous methods that simply combine object predictions from multiple feature maps from different network depths, the proposed network is equipped with embedded super-resolution layers (named as scale-transfer layer/module in this work) to explicitly explore the interscale consistency nature across multiple detection scales. Scale-transfer module naturally fits the base network with little computational cost. This module is further integrated with a dense convolutional network (DenseNet) to yield a one-stage object detector. We evaluate our proposed architecture on PASCAL VOC 2007 and MS COCO benchmark tasks and STDN obtains significant improvements over the comparable state-of-the-art detection models.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging

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

In Radiology, Nuclear Medicine and Imaging

Citations in the corpus

Bridging the Gap Between Anchor-Based and Anchor-Free Detection via Adaptive Training Sample Selection (2019, background): “Thereafter, plenty of works are presented to boost its performance in different aspects, such as fusing context information from different layers [24, 12, 69], training from scratch [50, 73], introducing new loss function [33, 6], anchor refinement and matching [66, 67], architecture redesign [21, 2”
EfficientDet: Scalable and Efficient Object Detection (2019, methodology): “Following this idea, PANet [19] adds an extra bottom-up path aggregation network on top of FPN; STDL [35] proposes a scale-transfer module to exploit cross-scale features; M2det [34] proposes a U-shape module to fuse multi-scale features, and G-FRNet [1] introduces gate units for controlling…”
NAS-FPN: Learning Scalable Feature Pyramid Architecture for Object Detection (2019, background): “Many recent works [7, 16, 17, 34, 38, 39, 40, 43, 41] propose various cross-scale connections or operations to combine features to generate pyramidal feature representations.”
Deep High-Resolution Representation Learning for Visual Recognition (2019, neutral): “result is from the original paper [9]. GFLOPs and #parameters of the models are given in Table7. backbone size LS AP AP 50 AP 75 AP S AP M AP L MLKP [116] VGG16 - - 28:6 52:4 31:6 10:8 33:4 45:1 STDN [149] DenseNet-169 513 - 31:8 51:0 33:6 14:4 36:1 43:4 DES [142] VGG16 512 - 32:8 53:2 34:6 13:9 36:”
Multi-Adversarial Faster-RCNN for Unrestricted Object Detection (2019, methodology): “In order to efficiently train the hierarchical domain feature alignment module, inspired by [41], we introduce a scale reduction module (SRM) which aims at down-scaling the feature maps without information loss.”
Deep Regionlets for Object Detection (2017, methodology): “ing superior performance over state-of-the-art. 2. Related Work Many approaches have been proposed for object detection including both traditional ones [13, 47, 45] and deep learning-based approaches [17, 40, 31, 38, 8, 16, 19, 9, 35, 6, 21, 55, 56, 54, 52, 46, 44]. Traditional approaches mainly use”
Object Detection in 20 Years: A Survey (2019, unknown): “Bottom-up fusion feeds forward shallow features to deeper layers via skip connections [237, 240–242].”
M2Det: A Single-Shot Object Detector based on Multi-Level Feature Pyramid Network (2018, methodology): “1, SSD (Liu et al. 2016) directly and independently uses two layers of the backbone ( i.e. VGG16) and four extra layers obtained by stride 2 convolution to construct the feature pyramid; STDN (Zhou et al. 2018) only uses the last dense block of DenseNet (Huang et al. 2017) to construct feature pyram”
Deep Learning for Generic Object Detection: A Survey (2018, background): “15 DenseNet201 [118] 18 200 − [321] Concatenate each layer with every other layer in a feed forward fashion.”
DetectoRS: Detecting Objects with Recursive Feature Pyramid and Switchable Atrous Convolution (2020, background): “PANet [49] adds another bottom-up path on top of FPN. STDL [82] proposes to exploit cross-scale features by a scale-transfer module.”
Recent Advances in Deep Learning for Object Detection (2019, background): “Thus, handling feature were later developed [109, 110, 109, 111, 112, 92, 113, 114, 115, 116, 117, 118, 119], with modiﬁcations to the feature pyramid block (see Fig.”
DC-SPP-YOLO: Dense Connection and Spatial Pyramid Pooling Based YOLO for Object Detection (2019, methodology): “5 Titan X STDN 300[21] 2018 DenseNet-169 78.”
Extended Feature Pyramid Network for Small Object Detection (2020, background): “STDN [30] employs sub-pixel convolution on top layers of DenseNet [31] to detect small objects and meanwhile reduce network parameters, but it is based on restricted information from a single feature map.”
MPDIoU: A Loss for Efficient and Accurate Bounding Box Regression (2023, methodology): “In deep models for object detection, R-CNN series [15], [16], [17] adopts two or three bounding box regression modules to obtain higher localization accuracy, while YOLO series [2, 3, 6] and SSD series [18, 19, 20] adopt one to achieve faster inference.”
Adaptive Convolution for Object Detection (2019, background): “STDN [24] combines features from different layers by a novel scale-transfer module, produces a high-resolution feature by inserting low-resolution features from multiple layers and can well handle the scale problem.”
Towards Large-Scale Small Object Detection: Survey and Benchmarks (2022, methodology): “…Feature Fusion: Deep CNN architecture produces hierarchical feature maps at different spatial resolutions, in which low-level features describe ﬁner details along with more localization cues, while high-level features capture richer semantic information [13], [43], [76], [82], [83], [84].”