Efficient and Accurate Arbitrary-Shaped Text Detection With Pixel Aggregation Network

Abstract

Scene text detection, an important step of scene text reading systems, has witnessed rapid development with convolutional neural networks. Nonetheless, two main challenges still exist and hamper its deployment to real-world applications. The first problem is the trade-off between speed and accuracy. The second one is to model the arbitrary-shaped text instance. Recently, some methods have been proposed to tackle arbitrary-shaped text detection, but they rarely take the speed of the entire pipeline into consideration, which may fall short in practical applications. In this paper, we propose an efficient and accurate arbitrary-shaped text detector, termed Pixel Aggregation Network (PAN), which is equipped with a low computational-cost segmentation head and a learnable post-processing. More specifically, the segmentation head is made up of Feature Pyramid Enhancement Module (FPEM) and Feature Fusion Module (FFM). FPEM is a cascadable U-shaped module, which can introduce multi-level information to guide the better segmentation. FFM can gather the features given by the FPEMs of different depths into a final feature for segmentation. The learnable post-processing is implemented by Pixel Aggregation (PA), which can precisely aggregate text pixels by predicted similarity vectors. Experiments on several standard benchmarks validate the superiority of the proposed PAN. It is worth noting that our method can achieve a competitive F-measure of 79.9% at 84.2 FPS on CTW1500. * Authors contributed equally. † Corresponding author.10.7% better 4 × faster Figure 1. The performance and speed on curved text dataset CTW1500. PAN-640 is 10.7% better than CTD+TLOC, and PAN-320 is 4 times faster than EAST. * indicates the results from [31].

Exploring Deep Transfer Learning Techniques for Alzheimer's Dementia Detection

“…Images of each concept are quality-controlled and human-annotated. ImageNet pre-training has been widely used in various computer vision tasks, such as fine-grained image classification ( Cui et al, 2018 ; Fu et al, 2017 ; Russakovsky et al, 2015 ), object detection ( Redmon et al, 2016 ; He et al, 2017 ), and sense text detection ( Zhou et al, 2017 ; Wang et al, 2019b ).…”

Accurate Scene Text Detection Via Scale-Aware Data Augmentation and Shape Similarity Constraint

“…Comparisons On Total-Text: As shown in Table XI, the Recall of our proposed method outperforms the state-of-the-art methods that do not pre-train on the synthetic dataset SynthText [12]. Specifically, our model improves the Recall of 1.8% compared with the detector PAN [41] in a single-scale testing. When we utilize ResNet-101 pre-trained on OpenImage to initialize our model and test the model using multiple scales, our method also works better than other methods that use multi-scale testing strategies and pre-train on SynthText.…”

“…For the segmentation-based methods, some researchers directly segment the text regions from the entire input image. Instead of only performing semantic segmentation for each pixel, more excellent approaches are proposed to learn more attributes, such as learning the link relationship among pixels [39], predicting the text border [24], learning the geometry attributes [7], [8], [27]- [32] of each pixel, constructing text instance with the progressive scale expansion [9], pulling pixels of the same text and pushing pixels of different text instances [40], [41], and so on. Besides, some methods perform the segmentation only on the bounding box.…”

Feature Selection Module for CNN Based Object Detector

“…While it has a similar network architecture composed of a backbone, neck, and head as in YoloV3. The feature extract backbone uses in YoloV4 is CSPDarket, and its neck uses a two-way FPN structure, which called PAN [56], instead of traditional FPN to fusion the multi-layer feature maps. FSM was then applied into the PAN architecture of YoloV4.…”