Abstract: Machine learning is used for a neural network multi-attribute facial encoder and decoder. A facial image is obtained for processing on a neural network and is encoded into two or more orthogonal feature subspaces. The encoding is performed by a single, trained encoder. The encoder is a downsampling encoder, orthogonality of the feature subspaces is established using metrics, and orthogonality enables separability of the feature subspaces. Embeddings are generated for two or more attributes of the facial image, wherein the embeddings are generated using one or more copies of the single, trained encoder. The embeddings comprise a vector representation of the two or more attributes of the facial image. A neural network is trained for a multi-task objective, wherein the training is based on the embeddings. The embeddings replace and augment training images. The multi-task objective provides identification of the two or more attributes of the facial image.

Abstract: Machine learning is performed using synthetic data for neural network training using vectors. Facial images are obtained for a neural network training dataset. Facial elements from the facial images are encoded into vector representations of the facial elements. A generative adversarial network (GAN) generator is trained to provide one or more synthetic vectors based on the one or more vector representations, wherein the one or more synthetic vectors enable avoidance of discriminator detection in the GAN. The training a GAN further comprises determining a generator accuracy using the discriminator. The generator accuracy can enable a classifier, where the classifier comprises a multi-layer perceptron. Additional synthetic vectors are generated in the GAN, wherein the additional synthetic vectors avoid discriminator detection. A machine learning neural network is trained using the additional synthetic vectors.

Abstract: Disclosed techniques include in-vehicle drowsiness analysis using blink-rate. Video of an individual is obtained within a vehicle using an image capture device. The video is analyzed using one or more processors to detect a blink event based on a classifier for a blink that was determined. Using temporal analysis, the blink event is determined by identifying that eyes of the individual are closed for a frame in the video. Using the blink event and one or more other blink events, blink-rate information is determined using the one or more processors. Based on the blink-rate information, a drowsiness metric is calculated using the one or more processors. The vehicle is manipulated based on the drowsiness metric. A blink duration of the individual for the blink event is evaluated. The blink-rate information is compensated. The compensating is based on demographic information for the individual.

Abstract: Disclosed techniques include neural network architecture using encoder-decoder models. A facial image is obtained for processing on a neural network. The facial image includes unpaired facial image attributes. The facial image is processed through a first encoder-decoder pair and a second encoder-decoder pair. The first encoder-decoder pair decomposes a first image attribute subspace. The second encoder-decoder pair decomposes a second image attribute subspace. The first encoder-decoder pair outputs a transformation mask based on the first image attribute subspace. The second encoder-decoder pair outputs a second image transformation mask based on the second image attribute subspace. The first image transformation mask and the second image transformation mask are concatenated to enable downstream processing. The concatenated transformation masks are processed on a third encoder-decoder pair and a resulting image is output. The resulting image eliminates a paired training data requirement.

Abstract: Machine learning is performed using synthetic data for neural network training using vectors. Facial images are obtained for a neural network training dataset. Facial elements from the facial images are encoded into vector representations of the facial elements. A generative adversarial network (GAN) generator is trained to provide one or more synthetic vectors based on the one or more vector representations, wherein the one or more synthetic vectors enable avoidance of discriminator detection in the GAN. The training a GAN further comprises determining a generator accuracy using the discriminator. The generator accuracy can enable a classifier, where the classifier comprises a multi-layer perceptron. Additional synthetic vectors are generated in the GAN, wherein the additional synthetic vectors avoid discriminator detection. A machine learning neural network is trained using the additional synthetic vectors.

Abstract: Disclosed techniques include in-vehicle drowsiness analysis using blink-rate. Video of an individual is obtained within a vehicle using an image capture device. The video is analyzed using one or more processors to detect a blink event based on a classifier for a blink that was determined. Using temporal analysis, the blink event is determined by identifying that eyes of the individual are closed for a frame in the video. Using the blink event and one or more other blink events, blink-rate information is determined using the one or more processors. Based on the blink-rate information, a drowsiness metric is calculated using the one or more processors. The vehicle is manipulated based on the drowsiness metric. A blink duration of the individual for the blink event is evaluated. The blink-rate information is compensated. The compensating is based on demographic information for the individual.