#GazeMoDiff

Code for "GazeMoDiff: Gaze-guided Diffusion Model for Stochastic Human Motion Prediction"

Environment Setup

sh install.sh

DATA PROCESSING

1. Download the original mogaze dataset from https://humans-to-robots-motion.github.io/mogaze/ and Download the original gimo dataset from https://github.com/y-zheng18/GIMO.

2. For MoGaze:

python motion_generation/mogaze_code/mogaze_preprocessing.py

For GIMO:

python motion_generation/gimo_code/gimo_preprocessing.py

3. Obtain the multimodal indices.

For MoGaze:

python humanmac/get_multimodal.py --dataset mogaze_withcontext

For GIMO:

python humanmac/get_multimodal.py --dataset gimo_withcontext

The processed data should be placed in "./data"

Training

For MoGaze:

python main.py --cfg mogaze_withcontext --mode train

For GIMO:

python main.py --cfg gimo_withcontext --mode train

Visualization of Motion Prediction

For MoGaze:

python main.py --cfg mogaze_withcontext  --mode pred --vis_row 1 --vis_col 10 --ckpt MODEL_PATH --ckpt_gcn GCNMODEL_PATH

For GIMO:

python main.py --cfg gimo_withcontext  --mode pred --vis_row 1 --vis_col 10 --ckpt MODEL_PATH  --ckpt_gcn GCNMODEL_PATH

## Evaluation

Evaluate on MoGaze:

python main.py --cfg mogaze_withcontext --mode eval --ckpt MODEL_PATH --ckpt_gcn GCNMODEL_PATH

Evaluate on GIMO:

python main.py --cfg gimo_withcontext --mode eval --ckpt MODEL_PATH --ckpt_gcn GCNMODEL_PATH

Contact at: [email protected]