hysts / pytorch_mpiigaze

An unofficial PyTorch implementation of MPIIGaze and MPIIFaceGaze
MIT License
346 stars 85 forks source link

what is normalized_camera_params for? #18

Closed Yijun88 closed 4 years ago

Yijun88 commented 4 years ago

Hi hysts,

Great work for the pytorch implementation! I was trying to understand the normalization process from the code and i found there's a normalized_camera_params.yaml that helped with the eye patch image normalization. Can I ask where do you get those information and why the camera matrix is different in the normalized_camera_params compared with the sample_params.yaml ?

====normalized_camera_params.yaml======

image_width: 60 image_height: 36 camera_matrix: rows: 3 cols: 3 data: [960., 0., 30, 0., 960., 18., 0., 0., 1.] distortion_coefficients: rows: 1 cols: 5 data: [0., 0., 0., 0., 0.]

=================================

Best, Yijun

hysts commented 4 years ago

Hi, @Yijun88

You can find the explanation of the image normalization in section 4.2 of this paper. I used parameters described in the paper for MPIIGaze, but I don't remember where I found parameters for MPIIFaceGaze. (I remember googling about it, so maybe in another paper?)

sample_params.yaml is a sample file for camera parameters of your real camera, and the parameters in normalized_camera_params.yaml are for a predefined virtual camera.