FoodLogoDet-1500: A Dataset for Large-Scale Food Logo Detection via Multi-Scale Feature Decoupling Network

Abstract

Food logo detection plays an important role in the multimedia for its wide real-world applications, such as food recommendation of the self-service shop and infringement detection on e-commerce platforms. A large-scale food logo dataset is urgently needed for developing advanced food logo detection algorithms. However, there are no available food logo datasets with food brand information. To support efforts towards food logo detection, we introduce the dataset FoodLogoDet-1500, a new large-scale publicly available food logo dataset, which has 1,500 categories, about 100,000 images and about 150,000 manually annotated food logo objects. We describe the collection and annotation process of FoodLogoDet-1500, analyze its scale and diversity, and compare it with other logo datasets.To the best of our knowledge, FoodLogoDet-1500 is the first largest publicly available high-quality dataset for food logo detection. The challenge of food logo detection lies in the large-scale categories and similarities between food logo categories. For that, we propose a novel food logo detection method Multi-scale Feature Decoupling Network (MFDNet), which decouples classification and regression into two branches and focuses on the classification branch to solve the problem of distinguishing multiple food logo categories. Specifically, we introduce the feature offset module, which utilizes the deformation-learning for optimal classification offset and can effectively obtain the most representative features of classification in detection. In addition, we adopt a balanced feature pyramid in * Corresponding Author.

Trinity‐Yolo: High‐precision logo detection in the real world

“…We compared Trinity-Yolo with recent research results, and the experimental results are presented in Table 6. Specifically, we compared it with the Logo-Yolo, MFDNet [47] and SeeTek [48] frameworks on two datasets. Logo-Yolo is an improved framework based on Yolov3.…”

International Comparison on the Application of Information Technology in High School Mathematics Textbooks

“…In this subsection, we show the main results of CTDNet conducted on four datasets. To validate the generality of the In Table IV, more contrastive experiments with logodetection-oriented methods are conducted on FlickrLogos-32 and QMUL-OpenLogo, including Logo-Yolo [13], OSF-Logo [28], MFDNet [14] and DSFP-GA [29].…”

“…Early methods for logo detection generally relied on manual feature extraction techniques and traditional classification models. Recently, a series of deep logo detection methods as well as large-scale datasets have been proposed by exploiting the state-of-the-art object detection models [12]- [15]. Logo-Yolo [13] was proposed to solve imbalanced samples of logos, and a high-quality logo dataset LogoDet-3K was built.…”

“…Logo-Yolo [13] was proposed to solve imbalanced samples of logos, and a high-quality logo dataset LogoDet-3K was built. MFDNet [14] was designed to address the multiscale and similar logo difficulties in food logo detection, and a large dataset FoodLogoDet-1500 was constructed to solve data limitations. A cross-view learning method [12] provided ideas for logo detection.…”

“…To evaluate the effectiveness of the proposed CTDNet, we complete comprehensive experimental validation on four datasets. They include two small-scale datasets FlickrLogos-32 [18] and QMUL-OpenLogo [19], the medium-scale food dataset FoodLogoDet-1500 [14], and the large-scale dataset LogoDet-3K [13]. The detailed description of these datasets is shown in Table I.…”