SSH: Single Stage Headless Face Detector

By Mahyar Najibi, Pouya Samangouei, Ramalingam Chellappa, Larry S. Davis. IEEE International Conference on Computer Vision, 2017

444 citations recorded by the source corpus.

We introduce the Single Stage Headless (SSH) face detector. Unlike two stage proposal-classification detectors, SSH detects faces in a single stage directly from the early convolutional layers in a classification network. SSH is headless. That is, it is able to achieve state-of-the-art results while removing the “head” of its underlying classification network – i.e. all fully connected layers in the VGG-16 which contains a large number of parameters. Additionally, instead of relying on an image pyramid to detect faces with various scales, SSH is scale-invariant by design. We simultaneously detect faces with different scales in a single forward pass of the network, but from different layers. These properties make SSH fast and light-weight. Surprisingly, with a headless VGG-16, SSH beats the ResNet-101-based state-of-the-art on the WIDER dataset. Even though, unlike the current state-of-the-art, SSH does not use an image pyramid and is 5X faster. Moreover, if an image pyramid is deployed, our light-weight network achieves state-of-the-art on all subsets of the WIDER dataset, improving the AP by 2.5%. SSH also reaches state-of-the-art results on the FDDB and Pascal-Faces datasets while using a small input size, leading to a speed of 50 frames/second on a GPU.

View Paper: https://www.semanticscholar.org/paper/a896ddeb0d253739c9aaef7fc1f170a2ba8407d3

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Benchmarking Low-Light Image Enhancement and Beyond (2021, background): “2017), Single Stage Headless Face Detector (SSH) (Najibi et al. 2017), Selective Refinement Network (SRN) (Chi et al.”
SNIPER: Efficient Multi-Scale Training (2018, methodology): “In the past, many different methods have been proposed to understand the role of context [35, 1, 22] and scale [3, 34, 18, 23].”
Label Smoothing and Logit Squeezing: A Replacement for Adversarial Training? (2019, background): “Deep Neural Networks (DNNs) have enjoyed great success in many areas of computer vision, such as classiﬁcation (Krizhevsky et al., 2012), object detection (Girshick, 2015), and face recognition (Najibi et al., 2017).”
Expression, Affect, Action Unit Recognition: Aff-Wild2, Multi-Task Learning and ArcFace (2019, methodology): “Visual Modality Pre-Processing The SSH detector [30] based on the ResNet and trained on the WiderFace dataset [46] was used to extract face bounding boxes from all images.”
AutoFocus: Efficient Multi-Scale Inference (2018, background): “Recently, feature-pyramids have been extensively studied and employed in deep learning based object detectors as the representation provides a boost in accuracy without compromising speed [42, 40, 64, 9, 33, 41, 47, 22, 34, 38].”
Towards Large-Scale Small Object Detection: Survey and Benchmarks (2022, methodology): “SSH [75] combines scale-variant face detectors, each trained for a certain scale range, to form a strong multi-scale detector to handle the faces varying extremely in scales.”
An Analysis of Scale Invariance in Object Detection - SNIP (2017, methodology): “Methods like SDP [35], SSH [27] or MS-CNN [3], which make independent predictions at different layers, also ensure that smaller objects are trained on higher resolution layers (like conv3) while larger objects are trained on lower resolution layers (like conv5).”
CIAGAN: Conditional Identity Anonymization Generative Adversarial Networks (2020, methodology): “We perform detection using HOG [5] and SSH Detector [27].”
RetinaFace: Single-stage Dense Face Localisation in the Wild (2019, methodology): “The most recent state-of-the-art methods [36, 68, 49] focus on single-stage [30, 29] design which densely samples face locations and scales on feature pyramids [28], demonstrating promising performance and yielding faster speed compared to two-stage methods [43, 63, 8].”
DSFD: Dual Shot Face Detector (2018, unknown): “Speciﬁcally, SSD [20], MS-CNN [2], SSH [23], S3FD [39] predict boxes on multiple layers of feature hierarchy.”
Detecting small faces in the wild based on generative adversarial network and contextual information (2019, methodology): “SSH [48] includes a simple context module which models the contextual information by increasing the effective receptive ﬁeld on the convolutional layers.”
YOLO-FaceV2: A Scale and Occlusion Aware Face Detector (2022, methodology): “SSH [37] uses means of simple convolution layers to incorporates context by enlarging the window around the candidate proposals.”
Finding Tiny Faces in the Wild with Generative Adversarial Network (2018, methodology): “We compare the our method with the state-of-the-art face detectors [31, 33, 37, 10, 19, 35].”
Res2Net: A New Multi-Scale Backbone Architecture (2019, background): “Multi-scale feature representations of CNNs are of great importance to a number of vision tasks including object detection [43], face analysis [4], [41], edge detection [37], semantic segmentation [6], salient object detection [34], [65], and skeleton detection [67], boosting the model performance o”
Multi-task Generative Adversarial Network for Detecting Small Objects in the Wild (2020, methodology): “…the AP performance of our method against the some previous state-of-the-art face detectors (Yang et al. 2016; Zhang et al. 2016; Zhu et al. 2017b; Hu and Ramanan 2017; Najibi et al. 2017; Zhang et al. 2017b; Wang et al. 2017a,b; Zhu et al. 2018; Zhang et al. 2018a) on WIDER FACE validation set.”
Recent Advances in Deep Learning for Object Detection (2019, background): “Single Stage Headless Face Detector (SSH) [189] was”
Analysing Affective Behavior in the First ABAW 2020 Competition (2020, methodology): “Then, the SSH detector [27] based on the ResNet [11] and trained on the WiderFace dataset [34] was used to extract face bounding boxes from all the images.”
MetaAnchor: Learning to Detect Objects with Customized Anchors (2018, background): “class labels, which leads to the following issues. First, anchor boxes need careful design, e.g. via clustering [30], which is especially critical on speciﬁc detection tasks such as anchor-based face [39, 44, 27, 35, 42] and pedestrian [40, 3, 43, 25] detections. Specially, some papers suggest multi”