The power requirements of video-oculography systems can be prohibitive for high-speed operation on portable devices. Recently, low-power alternatives such as photosensors have been evaluated, providing gaze estimates at high frequency with a trade-off in accuracy and robustness. Potentially, an approach combining slow/high-fidelity and fast/low-fidelity sensors should be able to exploit their complementarity to track fast eye motion accurately and robustly. To foster research on this topic, we introduce OpenSFEDS, a near-eye gaze estimation dataset containing approximately 2M synthetic camera-photosensor image pairs sampled at 500 Hz under varied appearance and camera position. We also formulate the task of sensor fusion for gaze estimation, proposing a deep learning framework consisting in appearance-based encoding and temporal eye-state dynamics. We evaluate several single- and multi-rate fusion baselines on OpenSFEDS, achieving 8.7% error decrease when tracking fast eye movements with a multi-rate approach vs. a gaze forecasting approach operating with a low-speed sensor alone.

Recently, image-to-image translation (I2I) has met with great success in computer vision, but few works have paid attention to the geometric changes that occur during translation. The geometric changes are necessary to reduce the geometric gap between domains at the cost of breaking correspondence between translated images and original ground truth. We propose a novel geometry-aware semi-supervised method to preserve this correspondence while still allowing geometric changes. The proposed method takes a synthetic image-mask pair as input and produces a corresponding real pair. We also utilize an objective function to ensure consistent geometric movement of the image and mask through the translation. Extensive experiments illustrate that our method yields a 11.23% higher mean Intersection-Over-Union than the current methods on the downstream eye segmentation task. The generated image has a 15.9% decrease in Frechet Inception Distance indicating higher image quality.

In this work, we investigate the hypothesis that adding 3D information of the periocular region to an end-to-end gaze-estimation network can improve gaze-estimation accuracy in the presence of slippage, which occurs quite commonly for head-mounted AR/VR devices. To this end, using UnityEyes we generate a simulated dataset with RGB and depth-maps of the eye with varying camera placement to simulate slippage artifacts. We generate different noise profiles for the depth-maps to simulate depth sensor noise artifacts. Using this data, we investigate the effects of different fusion techniques for combining image and depth information for gaze estimation. Our experiments show that under an attention-based fusion scheme, 3D information can significantly improve gaze-estimation and compensates well for slippage induced variability. Our finding supports augmenting 2D cameras with depth-sensors for the development of robust end-to-end appearance based gaze-estimation systems.

We investigate the hypothesis that gaze-signal can improve egocentric action recognition on the standard benchmark, EGTEA Gaze++ dataset. In contrast to prior work where gaze-signal was only used during training, we formulate a novel neural fusion approach, Cross-modality Attention Blocks (CMA), to leverage gaze-signal for action recognition during inference as well. CMA combines information from different modalities at different levels of abstraction to achieve state-of-the-art performance for egocentric action recognition. Specifically, fusing the video-stream with optical-flow with CMA outperforms the current state-of-the-art by 3%. However, when CMA is employed to fuse gaze-signal with video-stream data, no improvements are observed. Further investigation of this counter-intuitive finding indicates that small spatial overlap between the network’s attention-map and gaze ground-truth renders the gaze-signal uninformative for this benchmark. Based on our empirical findings, we recommend improvements to the current benchmark to develop practical systems for egocentric video understanding with gaze-signal.

Current methods for segmenting eye imagery into skin, sclera, pupil, and iris cannot leverage information about eye motion. This is because the datasets on which models are trained are limited to temporally non-contiguous frames. We present Temporal RIT-Eyes, a Blender pipeline that draws data from real eye videos for the rendering of synthetic imagery depicting natural gaze dynamics. These sequences are accompanied by ground-truth segmentation maps that may be used for training image-segmentation networks. Temporal RIT-Eyes relies on a novel method for the extraction of 3D eyelid pose (top and bottom apex of eyelids/eyeball boundary) from raw eye images for the rendering of gaze-dependent eyelid pose and blink behavior. The pipeline is parameterized to vary in appearance, eye/head/camera/illuminant geometry, and environment settings (indoor/outdoor). We present two open-source datasets of synthetic eye imagery: sGiW is a set of synthetic-image sequences whose dynamics are modeled on those of the Gaze in Wild dataset, and sOpenEDS2 is a series of temporally non-contiguous eye images that approximate the OpenEDS-2019 dataset. We also analyze and demonstrate the quality of the rendered dataset qualitatively and show significant overlap between latent-space representations of the source and the rendered datasets.

We present a large scale data set of eye-images captured using a virtual-reality (VR) head mounted display mounted with two synchronized eye-facing cameras at a frame rate of 200 Hz under controlled illumination. This dataset is compiled from video capture of the eye-region collected from 152 individual participants and is divided into four subsets: (i) 12,759 images with pixel-level annotations for key eye-regions: iris, pupil and sclera (ii) 252,690 unlabeled eye-images, (iii) 91,200 frames from randomly selected video sequences of 1.5 seconds in duration, and (iv) 143 pairs of left and right point cloud data compiled from corneal topography of eye regions collected from a subset, 143 out of 152, participants in the study. A baseline experiment has been evaluated on the dataset for the task of semantic segmentation of pupil, iris, sclera and background, with the mean intersection-over-union (mIoU) of 98.3 %. We anticipate that this dataset will create opportunities to researchers in the eye tracking community and the broader machine learning and computer vision community to advance the state of eye-tracking for VR applications, which in its turn will have greater implications in Human-Computer Interaction.

In recent years increasingly complex architectures for deep convolution networks (DCNs) have been proposed to boost the performance on image recognition tasks. However, the gains in performance have come at a cost of substantial increase in computation and model storage resources. Fixed point implementation of DCNs has the potential to alleviate some of these complexities and facilitate potential deployment on embedded hardware. In this paper, we propose a quantizer design for fixed point implementation of DCNs. We formulate and solve an optimization problem to identify optimal fixed point bit-width allocation across DCN layers. Our experiments show that in comparison to equal bit-width settings, the fixed point DCNs with optimized bit width allocation offer >20% reduction in the model size without any loss in accuracy on CIFAR-10 benchmark. We also demonstrate that fine-tuning can further enhance the accuracy of fixed point DCNs beyond that of the original floating point model. In doing so, we report a new state-of-the-art fixed point performance of 6.78% error-rate on CIFAR-10 benchmark.

Recently sequential model based optimization (SMBO) has emerged as a promising hyper-parameter optimization strategy in machine learning. In this work, we investigate SMBO to identify architecture hyper-parameters of deep convolution networks (DCNs) object recognition. We propose a simple SMBO strategy that starts from a set of random initial DCN architectures to generate new architectures, which on training perform well on a given dataset. Using the proposed SMBO strategy we are able to identify a number of DCN architectures that produce results that are comparable to state-of-the-art results on object recognition benchmarks.