FoveaBox: Beyound Anchor-Based Object Detection

By Tao Kong, Fuchun Sun, Huaping Liu, Yuning Jiang, Lei Li, Jianbo Shi. IEEE Transactions on Image Processing, 2019

774 citations recorded by the source corpus.

We present FoveaBox, an accurate, flexible, and completely anchor-free framework for object detection. While almost all state-of-the-art object detectors utilize predefined anchors to enumerate possible locations, scales and aspect ratios for the search of the objects, their performance and generalization ability are also limited to the design of anchors. Instead, FoveaBox directly learns the object existing possibility and the bounding box coordinates without anchor reference. This is achieved by: (a) predicting category-sensitive semantic maps for the object existing possibility, and (b) producing category-agnostic bounding box for each position that potentially contains an object. The scales of target boxes are naturally associated with feature pyramid representations. In FoveaBox, an instance is assigned to adjacent feature levels to make the model more accurate.We demonstrate its effectiveness on standard benchmarks and report extensive experimental analysis. Without bells and whistles, FoveaBox achieves state-of-the-art single model performance on the standard COCO and Pascal VOC object detection benchmark. More importantly, FoveaBox avoids all computation and hyper-parameters related to anchor boxes, which are often sensitive to the final detection performance. We believe the simple and effective approach will serve as a solid baseline and help ease future research for object detection. The code has been made publicly available at https://github.com/taokong/FoveaBox.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/0bd1de0668ba8fd4b4f9cab69950a0e5ea3453fb

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

Sketch-Guided Object Localization in Natural Images (2020, background): “State-of-the-art object localization methods can be primarily grouped into two categories: proposal-free [16,18,21,22,23,26] and proposalbased [1,7,8,10,11,24].”
IoU-balanced Loss Functions for Single-stage Object Detection (2019, methodology): “We evaluate the proposed IoU-balanced losses on the popular single-stage object detection models including anchor-based detectors(SSD[1], RetinaNet[3]) and anchor-free detector FoveaBox[23].”
VarifocalNet: An IoU-aware Dense Object Detector (2020, methodology): “Compared to the point feature used in most existing dense object detectors [7–9, 13 To verify the effectiveness of our proposed modules, we build the VFNet based on the FCOS and evaluate it on MS COCO benchmark [14].”
Sparse R-CNN: End-to-End Object Detection with Learnable Proposals (2020, background): “gorithms [16, 21, 36, 48, 19] are proposed to make this pipeline much simpler by replacing hand-crafted anchor boxes with reference points.”
Rethinking Transformer-based Set Prediction for Object Detection (2020, methodology): “We compare TSP-RCNN with multiple one-stage and two-stage object detection models [26, 31, 1, 28, 19, 40, 30, 3, 16, 35, 38] that also use ResNet-101 backbone or its deformable convolution network (DCN) [41] variant in Table 4.”
Detecting tiny objects in aerial images: A normalized Wasserstein distance and a new benchmark (2022, neutral): “…VGG-16 10.7 32.5 4.0 2.0 8.7 16.8 28.0 23.5 anchor-free: RepPonits (Yang et al., 2019) ResNet-50-FPN 9.3 23.6 5.4 2.8 10.0 12.3 18.9 20.0 FoveaBox (Kong et al., 2020) ResNet-50-FPN 11.3 28.1 7.4 1.4 8.6 17.8 32.2 21.8 FCOS (Tian et al., 2019) ResNet-50-FPN 12.0 30.2 7.3 2.2 11.1 16.6 26.9 26.0…”
PV-RCNN++: Point-Voxel Feature Set Abstraction With Local Vector Representation for 3D Object Detection (2021, background): “On the other hand, studies of the anchor-free direction mainly fall into keypoints-based [32], [33], [34], [35] and center-based [36], [37], [38], [39] paradigms.”
NAS-FCOS: Fast Neural Architecture Search for Object Detection (2019, background): “More recently, anchor-free one-stage detectors [9, 10, 24, 29, 30] have attracted increasing research interests, due to their simple fully convolutional architectures and reduced consumption of computational resources.”
A Survey of Deep Learning-Based Object Detection (2019, background): “To address that, recently propose a series of anchor-free methods [61] [62] [49] [108] [109] [110] [111] [112] [113]”
SOLOv2: Dynamic and Fast Instance Segmentation (2020, neutral): “2 FoveaBox [42] Res-101-FPN 40.”
Anchor-Free Oriented Proposal Generator for Object Detection (2021, background): “FCOS [40] and FoveaBox [41] directly regress the boxes from points.”
OTA: Optimal Transport Assignment for Object Detection (2021, background): “Without anchor box, FCOS [38], Foveabox [17] and their precursors [30, 14, 46] directly assign anchor points around the center of objects as positive samples, showing promising detection performance.”
RTM3D: Real-time Monocular 3D Detection from Object Keypoints for Autonomous Driving (2020, background): “The main architecture adopts a one-stage strategy that shares a similar layout with the anchor-free 2D object detector [38, 16, 47, 19], which allows us to get a fast detection speed.”