Can the accuracy bias by facial hairstyle be reduced through balancing the training data?

Appearance of a face can be greatly altered by growing a beard and mustache. The facial hairstyles in a pair of images can cause marked changes to the impostor distribution and the genuine distribution. Also, different distributions of facial hairstyle across demographics could cause a false impression of relative accuracy across demographics. We first show that, even though larger training sets boost the recognition accuracy on all facial hairstyles, accuracy variations caused by facial hairstyles persist regardless of the size of the training set. Then, we analyze the impact of having different fractions of the training data represent facial hairstyles. We created balanced training sets using a set of identities available in Webface42M that both have clean-shaven and facial hair images. We find that, even when a face recognition model is trained with a balanced clean-shaven / facial hair training set, accuracy variation on the test data does not diminish. Next, data augmentation is employed to further investigate the effect of facial hair distribution in training data by manipulating facial hair pixels with the help of facial landmark points and a facial hair segmentation model. Our results show facial hair causes an accuracy gap between clean-shaven and facial hair images, and this impact can be significantly different between African-Americans and Caucasians.

What is a Goldilocks Face Verification Test Set?

Face Recognition models are commonly trained with web-scraped datasets containing millions of images and evaluated on test sets emphasizing pose, age and mixed attributes. With train and test sets both assembled from web-scraped images, it is critical to ensure disjoint sets of identities between train and test sets. However, existing train and test sets have not considered this. Moreover, as accuracy levels become saturated, such as LFW $>99.8\%$, more challenging test sets are needed. We show that current train and test sets are generally not identity- or even image-disjoint, and that this results in an optimistic bias in the estimated accuracy. In addition, we show that identity-disjoint folds are important in the 10-fold cross-validation estimate of test accuracy. To better support continued advances in face recognition, we introduce two "Goldilocks" test sets, Hadrian and Eclipse. The former emphasizes challenging facial hairstyles and latter emphasizes challenging over- and under-exposure conditions. Images in both datasets are from a large, controlled-acquisition (not web-scraped) dataset, so they are identity- and image-disjoint with all popular training sets. Accuracy for these new test sets generally falls below that observed on LFW, CPLFW, CALFW, CFP-FP and AgeDB-30, showing that these datasets represent important dimensions for improvement of face recognition. The datasets are available at: \url{https://github.com/HaiyuWu/SOTA-Face-Recognition-Train-and-Test}

Identity Overlap Between Face Recognition Train/Test Data: Causing Optimistic Bias in Accuracy Measurement

A fundamental tenet of pattern recognition is that overlap between training and testing sets causes an optimistic accuracy estimate. Deep CNNs for face recognition are trained for N-way classification of the identities in the training set. Accuracy is commonly estimated as average 10-fold classification accuracy on image pairs from test sets such as LFW, CALFW, CPLFW, CFP-FP and AgeDB-30. Because train and test sets have been independently assembled, images and identities in any given test set may also be present in any given training set. In particular, our experiments reveal a surprising degree of identity and image overlap between the LFW family of test sets and the MS1MV2 training set. Our experiments also reveal identity label noise in MS1MV2. We compare accuracy achieved with same-size MS1MV2 subsets that are identity-disjoint and not identity-disjoint with LFW, to reveal the size of the optimistic bias. Using more challenging test sets from the LFW family, we find that the size of the optimistic bias is larger for more challenging test sets. Our results highlight the lack of and the need for identity-disjoint train and test methodology in face recognition research.

CRAFT: Contextual Re-Activation of Filters for face recognition Training

The first layer of a deep CNN backbone applies filters to an image to extract the basic features available to later layers. During training, some filters may go inactive, mean ing all weights in the filter approach zero. An inactive fil ter in the final model represents a missed opportunity to extract a useful feature. This phenomenon is especially prevalent in specialized CNNs such as for face recogni tion (as opposed to, e.g., ImageNet). For example, in one the most widely face recognition model (ArcFace), about half of the convolution filters in the first layer are inactive. We propose a novel approach designed and tested specif ically for face recognition networks, known as "CRAFT: Contextual Re-Activation of Filters for Face Recognition Training". CRAFT identifies inactive filters during training and reinitializes them based on the context of strong filters at that stage in training. We show that CRAFT reduces fraction of inactive filters from 44% to 32% on average and discovers filter patterns not found by standard training. Compared to standard training without reactivation, CRAFT demonstrates enhanced model accuracy on standard face-recognition benchmark datasets including AgeDB-30, CPLFW, LFW, CALFW, and CFP-FP, as well as on more challenging datasets like IJBB and IJBC.

LogicNet: A Logical Consistency Embedded Face Attribute Learning Network

Ensuring logical consistency in predictions is a crucial yet overlooked aspect in multi-attribute classification. We explore the potential reasons for this oversight and introduce two pressing challenges to the field: 1) How can we ensure that a model, when trained with data checked for logical consistency, yields predictions that are logically consistent? 2) How can we achieve the same with data that hasn't undergone logical consistency checks? Minimizing manual effort is also essential for enhancing automation. To address these challenges, we introduce two datasets, FH41K and CelebA-logic, and propose LogicNet, an adversarial training framework that learns the logical relationships between attributes. Accuracy of LogicNet surpasses that of the next-best approach by 23.05%, 9.96%, and 1.71% on FH37K, FH41K, and CelebA-logic, respectively. In real-world case analysis, our approach can achieve a reduction of more than 50% in the average number of failed cases compared to other methods.

Our Deep CNN Face Matchers Have Developed Achromatopsia

Modern deep CNN face matchers are trained on datasets containing color images. We show that such matchers achieve essentially the same accuracy on the grayscale or the color version of a set of test images. We then consider possible causes for deep CNN face matchers ``not seeing color''. Popular web-scraped face datasets actually have 30 to 60\% of their identities with one or more grayscale images. We analyze whether this grayscale element in the training set impacts the accuracy achieved, and conclude that it does not. Further, we show that even with a 100\% grayscale training set, comparable accuracy is achieved on color or grayscale test images. Then we show that the skin region of an individual's images in a web-scraped training set exhibit significant variation in their mapping to color space. This suggests that color, at least for web-scraped, in-the-wild face datasets, carries limited identity-related information for training state-of-the-art matchers. Finally, we verify that comparable accuracy is achieved from training using single-channel grayscale images, implying that a larger dataset can be used within the same memory limit, with a less computationally intensive early layer.

Beard Segmentation and Recognition Bias

A person's facial hairstyle, such as presence and size of beard, can significantly impact face recognition accuracy. There are publicly-available deep networks that achieve reasonable accuracy at binary attribute classification, such as beard / no beard, but few if any that segment the facial hair region. To investigate the effect of facial hair in a rigorous manner, we first created a set of fine-grained facial hair annotations to train a segmentation model and evaluate its accuracy across African-American and Caucasian face images. We then use our facial hair segmentations to categorize image pairs according to the degree of difference or similarity in the facial hairstyle. We find that the False Match Rate (FMR) for image pairs with different categories of facial hairstyle varies by a factor of over 10 for African-American males and over 25 for Caucasian males. To reduce the bias across image pairs with different facial hairstyles, we propose a scheme for adaptive thresholding based on facial hairstyle similarity. Evaluation on a subject-disjoint set of images shows that adaptive similarity thresholding based on facial hairstyles of the image pair reduces the ratio between the highest and lowest FMR across facial hairstyle categories for African-American from 10.7 to 1.8 and for Caucasians from 25.9 to 1.3. Facial hair annotations and facial hair segmentation model will be publicly available.

A Real Balanced Dataset For Understanding Bias? Factors That Impact Accuracy, Not Numbers of Identities and Images

The issue of disparities in face recognition accuracy across demographic groups has attracted increasing attention in recent years. Various face image datasets have been proposed as 'fair' or 'balanced' to assess the accuracy of face recognition algorithms across demographics. While these datasets often balance the number of identities and images across demographic groups. It is important to note that the number of identities and images in an evaluation dataset are not the driving factors for 1-to-1 face matching accuracy. Moreover, balancing the number of identities and images does not ensure balance in other factors known to impact accuracy, such as head pose, brightness, and image quality. We demonstrate these issues using several recently proposed datasets. To enhance the capacity for less biased evaluations, we propose a bias-aware toolkit that facilitates the creation of cross-demographic evaluation datasets balanced on factors mentioned in this paper.

Logical Consistency and Greater Descriptive Power for Facial Hair Attribute Learning

Face attribute research has so far used only simple binary attributes for facial hair; e.g., beard / no beard. We have created a new, more descriptive facial hair annotation scheme and applied it to create a new facial hair attribute dataset, FH37K. Face attribute research also so far has not dealt with logical consistency and completeness. For example, in prior research, an image might be classified as both having no beard and also having a goatee (a type of beard). We show that the test accuracy of previous classification methods on facial hair attribute classification drops significantly if logical consistency of classifications is enforced. We propose a logically consistent prediction loss, LCPLoss, to aid learning of logical consistency across attributes, and also a label compensation training strategy to eliminate the problem of no positive prediction across a set of related attributes. Using an attribute classifier trained on FH37K, we investigate how facial hair affects face recognition accuracy, including variation across demographics. Results show that similarity and difference in facial hairstyle have important effects on the impostor and genuine score distributions in face recognition.

Consistency and Accuracy of CelebA Attribute Values

We report the first analysis of the experimental foundations of facial attribute classification. An experiment with two annotators independently assigning values shows that only 12 of 40 commonly-used attributes are assigned values with >= 95% consistency, and that three (high cheekbones, pointed nose, oval face) have random consistency (50%). These results show that the binary face attributes currently used in this research area could re-focused to be more objective. We identify 5,068 duplicate face appearances in CelebA, the most widely used dataset in this research area, and find that individual attributes have contradicting values on from 10 to 860 of 5,068 duplicates. Manual audit of a subset of CelebA estimates error rates as high as 40% for (no beard=false), even though the labeling consistency experiment indicates that no beard could be assigned with >= 95% consistency. Selecting the mouth slightly open (MSO) attribute for deeper analysis, we estimate the error rate for (MSO=true) at about 20% and for (MSO=false) at about 2%. We create a corrected version of the MSO attribute values, and compare classification models created using the original versus corrected values. The corrected values enable a model that achieves higher accuracy than has been previously reported for MSO. Also, ScoreCAM visualizations show that the model created using the corrected attribute values is in fact more focused on the mouth region of the face. These results show that the error rate in the current CelebA attribute values should be reduced in order to enable learning of better models. The corrected attribute values for CelebA's MSO and the CelebA facial hair attributes will be made available upon publication.