Context-Anchors for Hybrid Resolution Face Detection

Wu, Tianpeng; Liang, Dong; Pan, Jiaxing; Kaneko, Shun’ichi

doi:10.1109/icip.2019.8803548

Cited by 8 publications

(9 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In our experiments, the models we used to verify our proposed methods are HR [9], EXTD [34], S 3 FD [35], LFFD [6], CAHR [30], PyramidBox [26], DSFD [12] and TinaFace [40]. All the models we used in the experiments are trained with the WIDER FACE training set and tested on the WIDER FACE validation set and Crowd Face.…”

Section: Experimental Settingmentioning

confidence: 99%

“…We introduce density information to the state-of-the-art anchor-based detectors, and then combine with our proposed algorithm. As is illustrated in Figure 4, we integrate FCP-DM to the trained detectors: HR [9], CAHR [30], EXTD [34], S 3 FD [35], Pyra-midBox [26] and DSFD [12], and compare their performance with the original detectors. The red curve in each figure represents our proposed co-occurrence prior based on the density map integrated into the detector.…”

Section: Experiments For Face Co-occurrence Prior Based On Density Mapmentioning

confidence: 99%

“…In this part, the score-size-specific NMS (S 3 NMS) is tested. The test sets include the WIDER FACE hard set and the Crowd Face, the test models include LFFD [6], HR [9], S3FD [35] , CAHR [30], PyramidBox [26], EXTD [34], DSFD [12] , the NMS threshold is 0.3.…”

Section: Experiments For Score-size-specific Nmsmentioning

confidence: 99%

See 2 more Smart Citations

Inferred box harmonization and aggregation for degraded face detection in crowds

Liang

Geng

Sun

et al. 2022

Multimed Tools Appl

Self Cite

View full text Add to dashboard Cite

Since objects usually keep a certain distance from the surveillance camera, small object detection is a practical issue. Detecting small objects is also one of the remaining challenges in the computer vision community. The current detectors usually leverage a more robust backbone network, build one or more multi-scale feature pyramids, or define a more precise anchor-box screening criteria. However, the distinguishable features are scarce due to the appearance degradation and a shallow resolution. In this paper, we leverage high-level context to enhance anchor-based detectors' capabilities for small and crowded face detection. We first define face cooccurrence prior based on density maps (FCP-DM) to explore extensive high-level contextual information. We propose a score-size-specific non-maximum suppression (S 3 NMS) to replace the traditional non-maximum suppression at the end of anchorbased detectors. Our approach is plug and play and model-independent, which could be concatenated into the existing anchor-based face detectors without extra learning. Compared to the prior art on the WIDER FACE hard set, our method increases an Average Precision of 0.1%-1.3%, while on Crowd Face, which we make for testing small and crowded face detection, it raises an Average Precision of 1% -6%. Codes and dataset have been available online.

show abstract

Section: Experimental Settingmentioning

confidence: 99%

Section: Experiments For Face Co-occurrence Prior Based On Density Mapmentioning

confidence: 99%

See 1 more Smart Citation

Inferred box harmonization and aggregation for degraded face detection in crowds

Liang

Geng

Sun

et al. 2022

Multimed Tools Appl

Self Cite

View full text Add to dashboard Cite

show abstract

“…We compared our approach with other post-processing methods NMS and Soft-NMS, which also do not need to re-train the model. We respectively integrate NMS, soft-NMS, and our proposed S 3 NMS into anchorbased detectors including HR [8], CAHR [32] and Pyra-midBox [28]. As shown in Table I, S 3 NMS has the highest Average Precision (AP) compared with NMS and soft-NMS on WIDER FACE hard set and Crowd Face set.…”

Section: B Experiments For Face Co-occurrence Prior Based On Density Mapmentioning

confidence: 99%

“…Different from lowlevel contextual information which adjusts the local receptive fields, our work extends the contextual information to the whole image rather than just surrounding objects. In our previous researches, we proposed a series of background modeling methods based on high-level contextual information to obtain stable context information between pixels for video foreground segmentation [11], [14], [15], [33], [32], [40], [34], [22], [12]. Inspired by this, in this paper, we try to introduce the high-level contextual information to hard face detection to improve the utilization efficiency of scene spatial information.…”

Section: Introductionmentioning

confidence: 99%