TinaFace: Strong but Simple Baseline for Face Detection

Zhu, Yanjia; Cai, Hongbin; Zhang, Shuhan; Wang, Chenhao; Xiong, Yi-song

doi:10.48550/arxiv.2011.13183

Cited by 26 publications

(44 citation statements)

References 62 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As presented in Table 5, the results are quite straight-forward: for regression branch, 𝛼 = 0.5 works better, while 𝛼 = 0. VGG-16 0.967 0.959 0.912 AInnoFace [53] ResNet-152 0.970 0.961 0.918 RetinaFace [8] ResNet-152 0.969 0.961 0.918 RefineFace [54] ResNet-152 0.972 0.962 0.920 DSFD [18] ResNet-152 0.966 0.957 0.904 ASFD-D6 [52] ResNet-152 0.972 0.965 0.925 HAMBox [25] ResNet-50 0.970 0.964 0.933 TinaFace [62] ResNet 7. Firstly, we test the Linear-Reweight proposed in [49], in which the factors that linearly increase from 1.0 to 2.0 are assigned to each FPN level.…”

Section: What Is the Best Label Assignment Disentanglement?mentioning

confidence: 99%

Disentangle Your Dense Object Detector

Chen

Yang

et al. 2021

Proceedings of the 29th ACM International Conference on Multimedia

View full text Add to dashboard Cite

Deep learning-based dense object detectors have achieved great success in the past few years and have been applied to numerous multimedia applications such as video understanding. However, the current training pipeline for dense detectors is compromised to lots of conjunctions that may not hold. In this paper, we investigate three such important conjunctions: 1) only samples assigned as positive in classification head are used to train the regression head; 2) classification and regression share the same input feature and computational fields defined by the parallel head architecture; and 3) samples distributed in different feature pyramid layers are treated equally when computing the loss. We first carry out a series of pilot experiments to show disentangling such conjunctions can lead to persistent performance improvement. Then, based on these findings, we propose Disentangled Dense Object Detector (DDOD), in which simple and effective disentanglement mechanisms are designed and integrated into the current state-of-the-art dense object detectors. Extensive experiments on MS COCO benchmark show that our approach can lead to 2.0 mAP, 2.4 mAP and 2.2 mAP absolute improvements on RetinaNet, FCOS, and ATSS baselines with negligible extra overhead. Notably, our best model reaches 55.0 mAP on the COCO test-dev set and 93.5 AP on the hard subset of WIDER FACE, achieving new state-of-the-art performance on these two competitive benchmarks. Code is available at https://github.com/zehuichen123/DDOD. CCS CONCEPTS• Computing methodologies → Object detection.

show abstract

Section: What Is the Best Label Assignment Disentanglement?mentioning

confidence: 99%

Disentangle Your Dense Object Detector

Chen

Yang

et al. 2021

Proceedings of the 29th ACM International Conference on Multimedia

View full text Add to dashboard Cite

show abstract

“…TinaFace This model was proposed to show that face detection is just a one class generic object detection problem -even "unique" characteristics of faces such as expression and makeup could correspond to distortion and color in objects [92]. At its core, TinaFace uses a ResNet-50 [39] with a Deformable Convolutional Network (DCN) [11] as the backbone and 6 level FPN [49] neck to extract the multi-scale features of the input image, an Inception block [75] to contextualize and broaden the receptive fields, a 5-layer Fully Convolutional Network (FCN) [53] classification head for classifying anchors, a 5-layer FCN regression head for bounding box regression of anchors, and a single convolutional layer that shares the first 4 convolutional layers with the regression head as the Intersection-of-Union-aware [82] (IoU-aware) head for predicting IoU of the face with the bounding box.…”

Section: Academic Modelsmentioning

confidence: 99%

Are Commercial Face Detection Models as Biased as Academic Models?

Dooley¹,

Wei²,

Goldstein³

et al. 2022

Preprint

View full text Add to dashboard Cite

As facial recognition systems are deployed more widely, scholars and activists have studied their biases and harms. Audits are commonly used to accomplish this and compare the algorithmic facial recognition systems' performance against datasets with various metadata labels about the subjects of the images. Seminal works have found discrepancies in performance by gender expression, age, perceived race, skin type, etc. These studies and audits often examine algorithms which fall into two categories: academic models or commercial models. We present a detailed comparison between academic and commercial face detection systems, specifically examining robustness to noise. We find that state-of-the-art academic face detection models exhibit demographic disparities in their noise robustness, specifically by having statistically significant decreased performance on older individuals and those who present their gender in a masculine manner. When we compare the size of these disparities to that of commercial models, we conclude that commercial models -in contrast to their relatively larger development budget and industry-level fairness commitments -are always as biased or more biased than an academic model. Recently, Dooley et al. [18] performed the first benchmark of commercial face detection systems' robustness to natural noise. They found that each of Amazon Rekognition, Microsoft Azure, and * Equal contribution.

show abstract

“…Following the pioneering work of Viola-Jones [32], numerous face detection algorithms have been designed. Among them, the single-shot anchor-based approaches [25,39,30,15,24,7,23,41] have recently demonstrated the most promising performance. In particular, on the most challenging face detection dataset WIDER FACE [34], the average precision (AP) on its Hard test set has been boosted to 92.4% by TinaFace [41].…”

Section: Introductionmentioning

confidence: 99%

“…InsightFace is a nonprofit face project. Even though TinaFace [41] achieves impressive results on unconstrained face detection, it employs large-scale (e.g. 1, 650 pixels) testing, which consumes huge amounts of computational cost (as shown in Tab.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Sample and Computation Redistribution for Efficient Face Detection

Guo,

Deng,

Lattas

et al. 2021

Preprint

View full text Add to dashboard Cite

Although tremendous strides have been made in uncontrolled face detection, efficient face detection with a low computation cost as well as high precision remains an open challenge. In this paper, we point out that training data sampling and computation distribution strategies are the keys to efficient and accurate face detection. Motivated by these observations, we introduce two simple but effective methods (1) Sample Redistribution (SR), which augments training samples for the most needed stages, based on the statistics of benchmark datasets; and (2) Computation Redistribution (CR), which reallocates the computation between the backbone, neck and head of the model, based on a meticulously defined search methodology. Extensive experiments conducted on WIDER FACE demonstrate the state-of-the-art efficiency-accuracy trade-off for the proposed SCRFD family across a wide range of compute regimes. In particular, SCRFD-34GF outperforms the best competitor, TinaFace, by 3.86% (AP at hard set) while being more than 3× faster on GPUs with VGA-resolution images. We also release our code to facilitate future research.

show abstract

TinaFace: Strong but Simple Baseline for Face Detection

Cited by 26 publications

References 62 publications

Disentangle Your Dense Object Detector

Disentangle Your Dense Object Detector

Are Commercial Face Detection Models as Biased as Academic Models?

Sample and Computation Redistribution for Efficient Face Detection

Contact Info

Product

Resources

About