LatentGaze: Cross-Domain Gaze Estimation through Gaze-Aware Analytic Latent Code Manipulation

This code is the PyTorch implementation of LatentGaze.

To prove our code's reproducibility, we present validation of LatentGaze on MPIIFaceGaze Datsets (9,000 images).

Due to encoding and generating the images tasks are very time-consuming, we prepare the gaze estimation code while except encoder-decoder codes. Hence, we prepare validation of LatentGaze in the single-domain task.

The encoder decoder code is being cleaned up. please wait

Datasets

Image data link : https://drive.google.com/file/d/1f_EugBGhJC9qZI1nOJq9mDI9YlnF50kG/view?usp=sharing Images dir : './dataset/MPII_validation'

Latent code link : https://drive.google.com/file/d/1ZGP8LFd0379ZznTv-2-cPLzyoCPW7KzD/view?usp=sharing latent codes dir : '/mpii_latent_pt_files_with_mpii/latent_pt_files'

LatentGaze weights

weights link : https://drive.google.com/file/d/19fNm8Pwnt-w_QSUMPoeLCH82SN2ZVD8p/view?usp=sharing

dir: './gaze_model_best.pt'

Create enviroments

conda env create -f LatentGaze.yaml

Quick Run

python main.py --test_only