Current appearance-based gaze estimation development mainly focuses on the frontal head pose. There are more challenging cases with extreme head poses in real-world scenarios, such as robot-human interaction, huge display in public places such as the airport and train station, room-level interaction where the camera can be placed in the corner, head-mounted cameras where the cameras are placed off to the side.
With our newly collected dataset, we are the first to enable to train a generic gaze estimation method with different head poses. We are aiming to learn the gaze mapping from input face image to the 3D gaze direction by feeding multi-view face images, and dynamically to adapt the head pose during the test. Such a model should learn the geometry of head poses as well as the corre-sponding feature representation of gaze direction.
For this project, we are aiming to submit the paper to the top-tier conference. Please send your application only if you are interested in publication. Thank you.
Current appearance-based gaze estimation development mainly focuses on the frontal head pose. There are more challenging cases with extreme head poses in real-world scenarios, such as robot-human interaction, huge display in public places such as the airport and train station, room-level interaction where the camera can be placed in the corner, head-mounted cameras where the cameras are placed off to the side.
With our newly collected dataset, we are the first to enable to train a generic gaze estimation method with different head poses. We are aiming to learn the gaze mapping from input face image to the 3D gaze direction by feeding multi-view face images, and dynamically to adapt the head pose during the test. Such a model should learn the geometry of head poses as well as the corre-sponding feature representation of gaze direction.
For this project, we are aiming to submit the paper to the top-tier conference. Please send your application only if you are interested in publication. Thank you.
Developing a gaze estimation method with multi-view images. Submitting a paper to a top-tier conference.
Developing a gaze estimation method with multi-view images. Submitting a paper to a top-tier conference.