search

eververas / 3DGazeNet

[ECCV 2024] 3DGazeNet: Generalizing Gaze Estimation with Weak-Supervision from Synthetic Views
https://eververas.github.io/3DGazeNet
62 stars 6 forks source link

for data from demo #15

Open ZhanXKun opened 2 months ago

ZhanXKun commented 2 months ago

Thank you very much for sharing, but when I extracted the data from the demo files, I found that the res18_x128_vert_xgz.pth required to run the demo was not present in the data.

eververas commented 2 months ago

Hello, we will soon update the demo code including this file path. For now you can try the model provided in the models download link in the main readme file.

ZhanXKun commented 2 months ago

Thank you very much for your response. Recently, I have tried loading the Gaze360 model and found that its gaze estimation is mostly based on the head orientation, which is not accurate enough. I believe the model you suggested will better suit my needs, and I am eagerly looking forward to your updates.

bisma-saleem commented 1 month ago

Kindly Update the Model.It will be a big help.Thank you

ZhanXKun commented 1 month ago

Kindly Update the Model.It will be a big help.Thank you

Have you tried using the data from the main readme file to process your own images? Do you think the output gaze vectors and other information are accurate? When I load my own video frames for processing, the gaze description in some images isn't very effective. T T