MuSe-2022 Baseline Model: LSTM Regressor

Sub-challenges and Results

For details, please see the Baseline Paper. If you want to sign up for the challenge, please fill the form here for MuSe-Humor and MuSe-Stress. For MuSe-Reaction, please contact competitions [at] hume.ai

MuSe-Humor: predicting presence/absence of humor in football press conference recordings. Official baseline: .8480 AUC.
MuSe-Reaction: predicting the intensity of seven emotions (Adoration, Amusement, Anxiety, Disgust, Empathic Pain, Fear, Surprise). Official baseline : .2801 mean Pearson's correlation over all seven classes.
MuSe-Stress: regression on valence and arousal signals for persons in a stressed disposition. Official baselines: .4931 CCC for valence, .4761 CCC for (physiological) arousal, .4585 CCC as the mean of best CCC for arousal and best CCC for valence (Combined). Note that the Combined score will be used to determine the challenge winner.

Installation

It is highly recommended to run everything in a Python virtual environment. Please make sure to install the packages listed in requirements.txt and adjust the paths in config.py (especially BASE_PATH).

You can then e.g. run the unimodal baseline reproduction calls in the *.sh file provided for each sub-challenge.

Settings

The main.py script is used for training and evaluating models. Most important options:

--task: choose either humor, reaction or stress
--feature: choose a feature set provided in the data (in the PATH_TO_FEATURES defined in config.py). Adding --normalize ensures normalization of features (recommended for eGeMAPS features).
Options defining the model architecture: d_rnn, rnn_n_layers, rnn_bi, d_fc_out
Options for the training process: --epochs, --lr, --seed, --n_seeds, --early_stopping_patience, --reduce_lr_patience, --rnn_dropout, --linear_dropout
In order to use a GPU, please add the flag --use_gpu
Specific parameters for MuSe-Stress: emo_dim (valence or physio-arousal), win_len and hop_len for segmentation.

For more details, please see the parse_args() method in main.py.

Reproducing the baselines

Unimodal results

For every challenge, a *.sh file is provided with the respective call (and, thus, configuration) for each of the precomputed features. Moreover, you can directly load one of the provided checkpoints corresponding to the results in the baseline paper. For MuSe-Humor, you can download the checkpoints here. The checkpoints for MuSe-Stress can be found here. Regarding MuSe-Reaction, the checkpoints are only available to registered participants. A checkpoint model can be loaded and evaluated as follows:

main.py --task humor --feature vggface2 --eval_model /your/checkpoint/directory/vggface2/model_102.pth

Note that egemaps features must be normalized (--normalize).

Fusion results

Late Fusion (MuSe-Humor, MuSe-Stress)

The idea of the late fusion implementation is to treat the predictions to be fused as a new feature set that is stored in the feature directory (alongside with e.g. egemaps, ds etc.). The script late_fusion.py creates such a feature set, given trained models. Example call:

python3 late_fusion_preparation.py --task humor --model_ids 2022-03-26-11-04_[ds]_[64_2_True_64]_[0.001_256] 2022-03-26-11-11_[vggface2]_[64_2_True_64]_[0.0001_256] --checkpoint_seeds 105 102

The model ids are directories under MODEL_FOLDER/task (MODEL_FOLDER is specified in config.py). For further details see the ArgumentParser in late_fusion_preparation.py.

This script will create a new feature set in the feature directory specified in config.py. This feature set can then be used with the main.py script as any other feature.

Early Fusion (MuSe-Reaction)

Similarly, for early fusion, an extra feature set is created. The corresponding script is early_fusion_preparation.py. Example call: python3 early_fusion_preparation.py --task reaction --feature_sets VGGFace2 DeepSpectrum

Citation:

The MuSe2022 baseline paper is available as a preprint here

@inproceedings{Christ22-TM2,
  title={The MuSe 2022 Multimodal Sentiment Analysis Challenge: Humor, Emotional Reactions, and Stress},
  author={Christ, Lukas and Amiriparian, Shahin and Baird, Alice and Tzirakis, Panagiotis and Kathan, Alexander and Müller, Niklas and Stappen, Lukas and Meßner, Eva-Maria and König, Andreas and Cowen, Alan and Cambria, Erik and Schuller, Bj\"orn W. },
  booktitle={Proceedings of the 3rd Multimodal Sentiment Analysis Challenge},
  year={2022},
  address = {Lisbon, Portugal},
  publisher = {Association for Computing Machinery},
  note = {co-located with ACM Multimedia 2022, to appear}
}

MuSe 2021 baseline paper:

@incollection{stappen2021muse,
  title={The MuSe 2021 multimodal sentiment analysis challenge: sentiment, emotion, physiological-emotion, and stress},
  author={Stappen, Lukas and Baird, Alice and Christ, Lukas and Schumann, Lea and Sertolli, Benjamin and Messner, Eva-Maria and Cambria, Erik and Zhao, Guoying and Schuller, Bj{\"o}rn W},
  booktitle={Proceedings of the 2nd on Multimodal Sentiment Analysis Challenge},
  pages={5--14},
  year={2021}
}

davidsjohnson/MuSe2022