Multistream Gaze Estimation with Anatomical Eye Region Isolation by Synthetic to Real Transfer Learning

TL;DR

A novel neural pipeline, MSGazeNet, that learns gaze representations by taking advantage of the eye anatomy information through a multistream framework is proposed and surpasses the state of the art by 7.57% and 1.85% on two datasets and obtains competitive results on the other.

Abstract

We propose a novel neural pipeline, MSGazeNet, that learns gaze representations by taking advantage of the eye anatomy information through a multistream framework. Our proposed solution comprises two components, first a network for isolating anatomical eye regions, and a second network for multistream gaze estimation. The eye region isolation is performed with a U-Net style network which we train using a synthetic dataset that contains eye region masks for the visible eyeball and the iris region. The synthetic dataset used in this stage is procured using the UnityEyes simulator, and consists of 80,000 eye images. Successive to training, the eye region isolation network is then transferred to the real domain for generating masks for the real-world eye images. In order to successfully make the transfer, we exploit domain randomization in the training process, which allows for the synthetic images to benefit from a larger variance with the help of augmentations that resemble artifacts. The generated eye region masks along with the raw eye images are then used together as a multistream input to our gaze estimation network, which consists of wide residual blocks. The output embeddings from these encoders are fused in the channel dimension before feeding into the gaze regression layers. We evaluate our framework on three gaze estimation datasets and achieve strong performances. Our method surpasses the state-of-the-art by 7.57% and 1.85% on two datasets, and obtains competitive results on the other. We also study the robustness of our method with respect to the noise in the data and demonstrate that our model is less sensitive to noisy data. Lastly, we perform a variety of experiments including ablation studies to evaluate the contribution of different components and design choices in our solution.

Figures (7)

• Figure 1: Overview of our proposed method. The anatomical eye region isolation module is used to generate binary eye region masks for the real-world eye images. These masks along with the raw eye image are then used by the multistream gaze estimator to perform gaze estimation. The pre-trained mask generation is represented via dotted lines and the solid lines represent the training pipeline.
• Figure 2: Our proposed framework, MSGazeNet. First, we perform anatomical eye region isolation using a U-Net style network which we train using the synthetic dataset. Next, we perform gaze estimation using real-world eye images and their corresponding eye region masks as input to our multistream gaze estimation network.
• Figure 3: The mask generation pipeline. The visible eyeball landmarks are marked in green dots and the iris landmarks are marked in red dots. The binary masks are extracted from the corresponding landmarks.
• Figure 4: The heatmap distribution of the gaze angles (top row) and the head pose angles (bottom row) of all the datasets.
• Figure 5: The outcome of the robustness analysis with respect to noise. Here we show the performance comparison of our proposed framework against Zhang et al. lenet, Zhang et al. mpii and Park et al. gazemap in the presence of different amounts of noise.