search

Ahmednull / L2CS-Net

The official PyTorch implementation of L2CS-Net for gaze estimation and tracking
MIT License
326 stars 80 forks source link

the best result #6

Open yuyyuu opened 2 years ago

yuyyuu commented 2 years ago

Hello, Thank for your great work. I have a question. When I replicated the work, I got a better result than 3.92 on MPIIFaceGaze. However, when I tested your pre-training results, I got a 4.17. I want to konw is 3.92 the best result ?Are all the results performed on 448 *448 pictures?

AimersS commented 2 years ago

Hello, Thank for your great work. I have a question. When I replicated the work, I got a better result than 3.92 on MPIIFaceGaze. However, when I tested your pre-training results, I got a 4.17. I want to konw is 3.92 the best result ?Are all the results performed on 448 *448 pictures?

Hello,When I trained his model, there were many bugs, and based on some values that were not defined, after modification, I got the wrong Mae: 0.041. How do you reproduce it? Can you analyze your code with me? My email: a847990008@icloud.com Thank you very much.

AndrewDing616 commented 2 years ago

Hello, when I replicated the work as the parameters the writer provided, I got 4.9°, where the error is big. I don't know why. So, could I have a communication with you? My email: 726804161@shu.edu.cn Tks!

godkaieethu commented 3 months ago

same question, bro. i get much smaller error on MpiiGaze