WIDER Face and Pedestrian Challenge 2018: Methods and Results

Loy, Chen Change; Lin, Dahua; Ouyang, Wanli; Xiong, Yuanjun; Yang, Shuo; Huang, Qingqiu; Zhou, Dongzhan; Xia, Wei; Li, Quanquan; Luo, Ping; Yan, Junjie; Wang, Jianfeng; Li, Zuoxin; Yuan, Ye; Li, Boxun; Shao, Shuai; Yu, Gang; Wei, Fangyun; Xiang, Ming; Chen, Dong; Zhang, Shifeng; Chi, Cheng; Lei, Zhen; Li, Stan Z.; Zhang, Hongkai; Ma, Bingpeng; Chang, Hong; Shan, Shiguang; Chen, Xilin; Liu, Wu; Zhou, Boyan; Li, Huaxiong; Cheng, Peng; Mei, Tao; Kukharenko, Artem; Vasenin, Artem; Sergievskiy, Nikolay; Yang, Hua; Li, Liangqi; Xu, Qiling; Yuan, Hong; Chen, Lin; Sun, Mingjian; Mao, Yirong; Luo, Sheng; Li, Yongjun; Wang, Ruiping; Xie, Qiaokang; Wu, Ziyang; Lü, Lei; Liu, Yiheng; Zhou, Wengang

doi:10.48550/arxiv.1902.06854

Cited by 12 publications

(24 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Although ECP [4] has a much larger scale, it still suffers from the low density of persons and high similarity of background scenes, which could be the focus of future datasets. Thus, in this work, we argue that the low density and diversity of these datasets constrains the generalization ability of pedestrian detectors, while the web crawled datasets, such as CrowdHuman [39], WiderPerson [54] and Wider Pedestrian [31], including much diverse scenes and denser persons may increase the upper bound of pedestrian detectors' generalization ability. Cross-dataset evaluation.…”

Section: Related Workmentioning

confidence: 87%

“…These three benchmarks include Caltech [13], CityPersons [52] and EuroCity Persons [4], and are categorized into the autonomous driving datasets in this work. Caltech [13] [31]. We provide more details of above datasets in Table 2.…”

Section: Experimental Settingsmentioning

confidence: 99%

“…2: Illustration of benchmarks. Top row shows images from diverse and dense datasets, such as CrowdHuman [39] and Wider Pedestrian [31]. Bottom row presents images of autonomous driving benchmarks, ECP [4], CityPersons [52] Caltech [13] We present results for Caltech and CityPersons in Table 5, respectively.…”

Section: Cross Dataset Evaluation Of Existing State-of-the-artmentioning

confidence: 99%

“…CrowdHuman [39] and Wider Pedestrian [31] datasets are two diverse and dense pedestrian detection datasets collected from web-crawling and surveillance cameras. Unlike the autonomous driving datasets, the crowd density and scenario diversities are large in CrowdHuman [39] and Wider Pedestrian [31] since they include images in diverse sources, such as street views and surveillance, which increases the data diversity from a different form. Thus, they are idea sources to pre-train models.…”

Section: Diverse General Person Detection Datasets For Generalizationmentioning

confidence: 99%

“…In recent years, few large and diverse person detection datasets, e.g. CrowdHuman [39], WiderPerson [54] and Wider Pedestrian [31], have been created using images and videos available online and through surveillance cameras. These datasets advance the general person detection research significantly but are not the most suitable datasets for pedestrian detection, as they contain people in a lot more diverse scenarios, than are relevant for autonomous driving.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Pedestrian Detection: Domain Generalization, CNNs, Transformers and Beyond

Hasan¹,

Liao²,

Li³

et al. 2022

Preprint

View full text Add to dashboard Cite

Pedestrian detection is the cornerstone of many vision based applications, starting from object tracking to video surveillance and more recently, autonomous driving. With the rapid development of deep learning in object detection, pedestrian detection has achieved very good performance in traditional single-dataset training and evaluation setting. However, in this study on generalizable pedestrian detectors, we show that, current pedestrian detectors poorly handle even small domain shifts in cross-dataset evaluation. We attribute the limited generalization to two main factors, the method and the current sources of data. Regarding the method, we illustrate that biasness present in the design choices (e.g anchor settings) of current pedestrian detectors are the main contributing factor to the limited generalization. Most modern pedestrian detectors are tailored towards target dataset, where they do achieve high performance in traditional single training and testing pipeline, but suffer a degrade in performance when evaluated through cross-dataset evaluation. Consequently, a general object detector performs better in cross-dataset evaluation compared with state of the art pedestrian detectors, due to its generic design. As for the data, we show that the autonomous driving benchmarks are monotonous in nature, that is, they are not diverse in scenarios and dense in pedestrians. Therefore, benchmarks curated by crawling the web (which contain diverse and dense scenarios), are an efficient source of pre-training for providing a more robust representation. Accordingly, we propose a progressive fine-tuning strategy which improves generalization. Additionally, this work also investigate the recent Transformer Networks as backbones to test generalization. We demonstrate that as of now, CNNs outperform transformer networks in terms of generalization and absorbing large scale datasets for learning robust representation. In conclusion, this paper suggests a paradigm shift towards cross-dataset evaluation, for the next generation of pedestrian detectors. Code and models can be accessed at https://github.com/hasanirtiza/Pedestron.

show abstract

Section: Related Workmentioning

confidence: 87%

Section: Experimental Settingsmentioning

confidence: 99%

Section: Cross Dataset Evaluation Of Existing State-of-the-artmentioning

confidence: 99%

Section: Diverse General Person Detection Datasets For Generalizationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Pedestrian Detection: Domain Generalization, CNNs, Transformers and Beyond

Hasan¹,

Liao²,

Li³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

OmDet: Large‐scale vision‐language multi‐dataset pre‐training with multimodal detection network

Zhao,

Liu,

Lee

2024

IET Computer Vision

View full text Add to dashboard Cite

The advancement of object detection (OD) in open‐vocabulary and open‐world scenarios is a critical challenge in computer vision. OmDet, a novel language‐aware object detection architecture and an innovative training mechanism that harnesses continual learning and multi‐dataset vision‐language pre‐training is introduced. Leveraging natural language as a universal knowledge representation, OmDet accumulates “visual vocabularies” from diverse datasets, unifying the task as a language‐conditioned detection framework. The multimodal detection network (MDN) overcomes the challenges of multi‐dataset joint training and generalizes to numerous training datasets without manual label taxonomy merging. The authors demonstrate superior performance of OmDet over strong baselines in object detection in the wild, open‐vocabulary detection, and phrase grounding, achieving state‐of‐the‐art results. Ablation studies reveal the impact of scaling the pre‐training visual vocabulary, indicating a promising direction for further expansion to larger datasets. The effectiveness of our deep fusion approach is underscored by its ability to learn jointly from multiple datasets, enhancing performance through knowledge sharing.

show abstract

A Hybrid Approach for Semantic Image Annotation

2021

View full text Add to dashboard Cite

In this study, a framework is proposed that generates natural language descriptions of images within a controlled environment. Previous work on neural networks mostly focused on choosing the right labels and/or increasing the number of related labels to depict an image. However, creating a textual description of an image is a completely different phenomenon, structurally, syntactically, and semantically. The proposed semantic image annotation framework presents a novel combination of deep learning models and aligned annotation results derived from the instances of the ontology classes to generate sentential descriptions of images. Our hybrid approach benefits from the unique combination of deep learning and semantic web technologies. We detect objects from unlabeled sports images using a deep learning model based on a residual network and a feature pyramid network, with the focal loss technique to obtain predictions with high probability. The proposed framework not only produces probabilistically labeled images, but also the contextual results obtained from a knowledge base exploiting the relationship between the objects. The framework's object detection and prediction performances are tested with two datasets where the first one includes individual instances of images containing everyday scenes of common objects and the second custom dataset contains sports images collected from the web. Moreover, a sample image set is created to obtain annotation result data by applying all framework layers. Experimental results show that the framework is effective in this controlled environment and can be used with other applications via web services within the supported sports domain.INDEX TERMS Semantic image annotation; picture interpretation; ontology.

show abstract

WIDER Face and Pedestrian Challenge 2018: Methods and Results

Cited by 12 publications

References 36 publications

Pedestrian Detection: Domain Generalization, CNNs, Transformers and Beyond

Pedestrian Detection: Domain Generalization, CNNs, Transformers and Beyond

OmDet: Large‐scale vision‐language multi‐dataset pre‐training with multimodal detection network

A Hybrid Approach for Semantic Image Annotation

Contact Info

Product

Resources

About