Search for: All records

Facial attribute prediction is a facial analysis task that describes images using natural language features. While many works have attempted to optimize prediction accuracy on CelebA, the largest and most widely used facial attribute dataset, few works have analyzed the accuracy of the dataset's attribute labels. In this paper, we seek to do just that. Despite the popularity of CelebA, we find through quantitative analysis that there are widespread inconsistencies and inaccuracies in its attribute labeling. We estimate that at least one third of all images have one or more incorrect labels, and reliable predictions are impossible for several attributes due to inconsistent labeling. Our results demonstrate that classifiers struggle with many CelebA attributes not because they are difficult to predict, but because they are poorly labeled. This indicates that the CelebA dataset is flawed as a facial analysis tool and may not be suitable as a generic evaluation benchmark for imbalanced classification. 

CelebA is the most common and largest scale dataset used to evaluate methods for facial attribute prediction, an important benchmark in imbalanced classification and face analysis. However, we argue that the evaluation metrics and baseline models currently used to compare the performance of different methods are insufficient for determining which approaches are best at classifying highly imbalanced attributes. We are able to obtain results comparable to current state-of-the-art using a ResNet-18 model trained with binary cross-entropy, a substantially less sophisticated approach than related work. We also show that we can obtain near-state-of-the-art results on accuracy using a model trained with just 10% of CelebA, and on balanced accuracy simply by maximizing recall for imbalanced attributes at the expense of all other metrics. To deal with these issues, we suggest several improvements to model evaluation including better metrics, stronger baselines, and increased awareness of the limitations of the dataset. 

One significant challenge in the field of supervised deep learning is the lack of large-scale labeled datasets for many problems. In this paper, we propose Consensus Spectral Clustering (CSC), which leverages the strengths of convolutional autoencoders and spectral clustering to provide pseudo labels for image data. This data can be used as weakly-labeled data for training and evaluating classifiers which require supervision. The primary weaknesses of previous works lies in their inability to isolate the object of interest in an image and cluster similar images together. We address these issues by denoising input images to remove pixels which do not contain data pertinent to the target. Additionally, we introduce a voting method for label selection to improve the clustering results. Our extensive experimentation on several benchmark datasets demonstrates that the proposed CSC method achieves competitive performance with state-of-the-art methods. 

We introduce a novel algorithm – ConvNEAT – that evolves a convolutional neural network (CNN) from a minimal architecture. Convolutional and dense nodes are evolved without restriction to the number of nodes or connections between nodes. The proposed work advances the field with ConvNEAT’s ability to evolve arbitrary minimal architectures with multi-dimensional inputs using GPU processing.