Each participant team should submit the following documents:
Python-based source code;
Checkpoint of the final model;
A detailed and accurate README file (Ensure that the code can be run successfully according to provided README file);
Results achieved on validation and test sets.
Submission via email to: s.song@exeter.ac.uk
Submission deadline: July 5th 2025 (Anywhere of Earth)
=======================================================================================
1. Source code
Please include all parts of the system, including training scripts, evaluation scripts, models, pre-processing scripts, etc.
We encourage participants to keep their codes having a similar environment and style as our baseline code provided at https://github.com/reactmultimodalchallenge/baseline_react2025. Alternatively, participants should submit a docker file including the code and the corresponding specific environment.
The data directory must be consistent with the baseline code (please refer to the file ‘data.py’ provided by the baseline).
We encourage participants to use 3DMM and PIRender-based strategy for facial reaction visualization. Please provide the details of the visualization if other strategy is applied.
2. Model and checkpoint
Please provide the following model definition python file and the corresponding checkpoints:
(1) Generation model for Facial Attributes, i.e., the occurrence (0 or 1) of 15 facial action units (i.e., AU1, AU2, AU4, AU6, AU7, AU9, AU10, AU12, AU14, AU15, AU17, AU23, AU24, AU25 and AU26), 2 facial affects - valence and arousal intensities (range from -1 to 1) - and the probabilities (range from 0 to 1) of eight categorical facial expressions (Neutral, Happy, Sad, Surprise, Fear, Disgust, Anger and Contempt), utilized for metric evaluation.
(2) Facial reaction 2D-Videos (frame-wise sequences) / 3D-Animation (3DMM-coefficients sequences) generation model, typically utilized for visualization.
3. README file
(1) Indicating the model type: offline facial reaction generation model or online facial reaction generation model.
(2) Detailed description of the coding environment and dependencies.
(3) Detailed description of the input and output:
(3.1) Sequences of facial attributes (AUs);
(3.2) 2D frame sequences (i.e., frame-wise video);
(3.3) 3D facial reaction representations (3DMM coefficients).
(4) Detailed description of dataloader, pre-processing and training settings (including training steps, hyper-parameters, etc.)
(5) Detailed description of visualization settings, including how to generate
(5.1) Either 2D frame-wise sequences (2D video);
(5.2) Or Sequences of 3DMM coefficients (3D animation).
(6) Indicating the data directory. The organizer will directly replace the ‘data directory’ in the submitted code with our data directory, and thus the data file directory must be consistent with the baseline code.
(7) Clearly list several ‘command lines’, where these command lines should allow organizer to:
(7.1) conduct model training;
(7.2) load pre-trained weights and evaluate it on validation/test set;
(7.3) generate 25-dimension facial attributes and 2D frame-wise sequences / 3DMM coefficients sequences.
4. Results achieved on validation and test sets
Coming soon ...