Rotation-Sensitive Regression for Oriented Scene Text Detection

By Minghui Liao, Zhen Zhu, Baoguang Shi, Gui-Song Xia, Xiang Bai. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018

487 citations recorded by the source corpus.

Text in natural images is of arbitrary orientations, requiring detection in terms of oriented bounding boxes. Normally, a multi-oriented text detector often involves two key tasks: 1) text presence detection, which is a classification problem disregarding text orientation; 2) oriented bounding box regression, which concerns about text orientation. Previous methods rely on shared features for both tasks, resulting in degraded performance due to the incompatibility of the two tasks. To address this issue, we propose to perform classification and regression on features of different characteristics, extracted by two network branches of different designs. Concretely, the regression branch extracts rotation-sensitive features by actively rotating the convolutional filters, while the classification branch extracts rotation-invariant features by pooling the rotation-sensitive features. The proposed method named Rotation-sensitive Regression Detector (RRD) achieves state-of-the-art performance on several oriented scene text benchmark datasets, including ICDAR 2015, MSRA-TD500, RCTW-17, and COCO-Text. Furthermore, RRD achieves a significant improvement on a ship collection dataset, demonstrating its generality on oriented object detection.

Subfields: Computer Vision and Pattern Recognition, Media Technology

View Paper: https://www.semanticscholar.org/paper/31f8c58679b52ff5c67bf4e46704c678a2968d6f

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

1. Dacheng Tao
2. Luc Van Gool
3. Chunhua Shen
4. Xiang Bai
5. Qi Tian
6. Junchi Yan
7. Ming-Hsuan Yang
8. Yi Yang

In Media Technology

1. Cong Yao
2. Xiang Bai
3. Minghui Liao
4. Xin He
5. Xue Yang
6. Shangbang Long
7. Junchi Yan
8. Guisong Xia

Citations in the corpus

• Align Deep Features for Oriented Object Detection (2020, methodology): “Inspired by [14] and [30], we first adopt ARF to encode the orientation information and then extract orientation-sensitive features and orientation-invariant features for regression and classification, respectively.”
• RBox-CNN: rotated bounding box based CNN for ship detection in remote sensing image (2018, methodology): “Moreover, compared with the rotation-sensitive regression detector (RRD) [3], our model also performs 7.6% better on L1 task.”
• Arbitrary Shape Scene Text Detection With Adaptive Text Region Representation (2019, methodology): “Bounding box regression based meth-ods [5, 8, 11, 12, 13, 16], which are inspired by general object detection methods such as SSD [14] and Faster R-CNN [23], treat text as a kind of object and directly estimate its bounding box as the detection result.”
• ReDet: A Rotation-equivariant Detector for Aerial Object Detection (2021, unknown): “method RC2 [19] RRPN [22] R(2)PN [43] RRD [16] RoI Trans.”
• Learning Modulated Loss for Rotated Object Detection (2019, methodology): “And then we resort to the eight-parameter regression model [22, 23, 24, 25] (As will be shown later in the paper, the discontinuity caused by rotation boundary inherently exist in eight-parameter system) to avoid regression parameter inconsistency.”
• Rethinking Rotated Object Detection with Gaussian Wasserstein Distance Loss (2021, background): “Arbitrary-oriented objects are ubiquitous for detection across visual datasets, such as aerial images [65, 2, 8, 69], scene text [82, 31, 20, 35, 27], faces [47] and 3D objects [78], retail scenes [5, 41], etc.”
• Character Region Awareness for Text Detection (2019, unknown): “In recent, Rotation-Sensitive Regression Detector (RSDD) [19] which makes full use of rotation-invariant features by actively rotating the convolutional filters was proposed.”
• An Empirical Study of Remote Sensing Pretraining (2022, background): “NC Head [1, 1, 6] [1, 2, 4, 8] Group [1, 1, 96] [1, 32, 64, 128] Type [F, F, F] [W, W, F, F] Depth [0, 0, 14] [2, 2, 8, 2]”
• PIoU Loss: Towards Accurate Oriented Object Detection in Complex Environments (2020, methodology): “Existing OBB-based approaches are mostly built on anchor-based frameworks by introducing an additional angle dimension optimized by a distance loss [24,18,19,6,41,43] on the parameter tuple (cx, cy, w, h, θ).”
• Real-time Scene Text Detection with Differentiable Binarization (2019, methodology): “TextBoxes (Liao et al. 2017), TextBoxes++ (Liao, Shi, and Bai 2018), SegLink (Shi, Bai, and Belongie 2017), and RRD (Liao et al. 2018) achieved fast text detection by following the detection architecture of SSD (Liu et al. 2016).”
• Dynamic Anchor Learning for Arbitrary-Oriented Object Detection (2020, unknown): “ ResNet101 800 800 54 79.08 R2PN(Zhang et al. 2018) VGG16 - 24 79.60 RoI Trans. (Ding et al. 2019) ResNet101 512 800 5 86.20 Gliding Vertex(Xu et al. 2020) ResNet101 512 800 5 88.20 Single-stage: RRD(Liao et al. 2018) VGG16 384 384 13 84.30 R3Det(Yang et al. 2019a) ResNet101 800 800 21 89.26 R-Retin”
• Learning RoI Transformer for Oriented Object Detection in Aerial Images (2019, methodology): “Specifically, based on our proposed method, the mAP can reach 86.16, 1.86 higher than that of RRD [20].”
• SCRDet++: Detecting Small, Cluttered and Rotated Objects via Instance-Level Feature Denoising and Rotation Loss Smoothing (2020, background): “TextBoxes++ [43] adopts vertex regression on SSD. RRD [44] further improves TextBoxes++ by decoupling classification and bounding box regression on rotation-invariant and rotation sensitive features, respectively.”
• Look More Than Once: An Accurate Detector for Text of Arbitrary Shapes (2019, methodology): “Detection-based methods [14, 15, 24, 42, 8] usually adopt some popular object detection frameworks and models under the supervision of the word or line level annotations.”
• Dynamic Refinement Network for Oriented and Densely Packed Object Detection (2020, unknown): “Method CP [28] BL2 [28] RC1 [28] RC2 [28] R (2)PN [43] RRD [21] RoI Trans [5] Ours”
• Efficient and Accurate Arbitrary-Shaped Text Detection With Pixel Aggregation Network (2019, background): “RRD [28] applies rotation-invariant and sensitive features for text classification and regression from two separate branches for better long text detection.”
• Arbitrary-Oriented Object Detection with Circular Smooth Label (2020, methodology): “31 70.44 73.51 77.62 90.84 86.15 86.69 69.60 68.04 73.83 71.1068.9376.17 Table 6. Detection accuracy on HRSC2016 dataset. Method R2CNN [12] RC1 & RC2 [26] RRPN [27] R2PN [46] RetinaNet-H [41] RRD [19] mAP (07) 73.07 75.7 79.08 79.6 82.89 84.30 Method RoI-Transformer [4] RSDet [30] Gliding Vertex”