SimMMDG: A Simple and Effective Framework for Multi-modal Domain Generalization
1ETH Zurich, 2EPFL
---
Overview of SimMMDG. We split the features of each modality into modality-specific and modality-shared parts. For the modality-shared part, we use supervised contrastive learning to map the features with the same label to be as close as possible. For modality-specific features, we use a distance loss to encourage them to be far from modality-shared features, promoting diversity within each modality. Additionally, we introduce a cross-modal translation module that regularizes features and enhances generalization across missing modalities.
Code
The code was tested using Python 3.10.4
, torch 1.11.0+cu113
and NVIDIA GeForce RTX 3090
.
Environments:
mmcv-full 1.2.7
mmaction2 0.13.0
EPIC-Kitchens Dataset
Prepare
Download Pretrained Weights
-
Download Audio model link, rename it as vggsound_avgpool.pth.tar
and place under the EPIC-rgb-flow-audio/pretrained_models
directory
-
Download SlowFast model for RGB modality link and place under the EPIC-rgb-flow-audio/pretrained_models
directory
-
Download SlowOnly model for Flow modality link and place under the EPIC-rgb-flow-audio/pretrained_models
directory
Download EPIC-Kitchens Dataset
bash download_script.sh
Download Audio files EPIC-KITCHENS-audio.zip.
Unzip all files and the directory structure should be modified to match:
Click for details...
```
├── MM-SADA_Domain_Adaptation_Splits
├── rgb
| ├── train
| | ├── D1
| | | ├── P08_01.wav
| | | ├── P08_01
| | | | ├── frame_0000000000.jpg
| | | | ├── ...
| | | ├── P08_02.wav
| | | ├── P08_02
| | | ├── ...
| | ├── D2
| | ├── D3
| ├── test
| | ├── D1
| | ├── D2
| | ├── D3
├── flow
| ├── train
| | ├── D1
| | | ├── P08_01
| | | | ├── u
| | | | | ├── frame_0000000000.jpg
| | | | | ├── ...
| | | | ├── v
| | | ├── P08_02
| | | ├── ...
| | ├── D2
| | ├── D3
| ├── test
| | ├── D1
| | ├── D2
| | ├── D3
```
Video and Audio
Click for details...
```
cd EPIC-rgb-flow-audio
```
```
python train_video_flow_audio_EPIC_SimMMDG.py --use_video --use_audio -s D2 D3 -t D1 --lr 1e-4 --bsz 16 --nepochs 20 --datapath /path/to/EPIC-KITCHENS/
```
```
python train_video_flow_audio_EPIC_SimMMDG.py --use_video --use_audio -s D1 D3 -t D2 --lr 1e-4 --bsz 16 --nepochs 20 --datapath /path/to/EPIC-KITCHENS/
```
```
python train_video_flow_audio_EPIC_SimMMDG.py --use_video --use_audio -s D1 D2 -t D3 --lr 1e-4 --bsz 16 --nepochs 25 --datapath /path/to/EPIC-KITCHENS/
```
Video and Flow
Click for details...
```
cd EPIC-rgb-flow-audio
```
```
python train_video_flow_audio_EPIC_SimMMDG.py --use_video --use_flow -s D2 D3 -t D1 --lr 1e-4 --bsz 16 --nepochs 15 --datapath /path/to/EPIC-KITCHENS/
```
```
python train_video_flow_audio_EPIC_SimMMDG.py --use_video --use_flow -s D1 D3 -t D2 --lr 1e-4 --bsz 16 --nepochs 15 --datapath /path/to/EPIC-KITCHENS/
```
```
python train_video_flow_audio_EPIC_SimMMDG.py --use_video --use_flow -s D1 D2 -t D3 --lr 1e-4 --bsz 16 --nepochs 15 --datapath /path/to/EPIC-KITCHENS/
```
Flow and Audio
Click for details...
```
cd EPIC-rgb-flow-audio
```
```
python train_video_flow_audio_EPIC_SimMMDG.py --use_flow --use_audio -s D2 D3 -t D1 --lr 1e-4 --bsz 16 --nepochs 10 --datapath /path/to/EPIC-KITCHENS/
```
```
python train_video_flow_audio_EPIC_SimMMDG.py --use_flow --use_audio -s D1 D3 -t D2 --lr 1e-4 --bsz 16 --nepochs 20 --datapath /path/to/EPIC-KITCHENS/
```
```
python train_video_flow_audio_EPIC_SimMMDG.py --use_flow --use_audio -s D1 D2 -t D3 --lr 1e-4 --bsz 16 --nepochs 20 --datapath /path/to/EPIC-KITCHENS/
```
Video and Flow and Audio
Click for details...
```
cd EPIC-rgb-flow-audio
```
```
python train_video_flow_audio_EPIC_SimMMDG.py --use_video --use_flow --use_audio -s D2 D3 -t D1 --lr 1e-4 --bsz 16 --nepochs 10 --trans_hidden_num 1024 --datapath /path/to/EPIC-KITCHENS/
```
```
python train_video_flow_audio_EPIC_SimMMDG.py --use_video --use_flow --use_audio -s D1 D3 -t D2 --lr 1e-4 --bsz 16 --nepochs 20 --datapath /path/to/EPIC-KITCHENS/
```
```
python train_video_flow_audio_EPIC_SimMMDG.py --use_video --use_flow --use_audio -s D1 D2 -t D3 --lr 1e-4 --bsz 16 --nepochs 15 --alpha_trans 1.0 --datapath /path/to/EPIC-KITCHENS/
```
HAC Dataset
This dataset can be downloaded at link.
Unzip all files and the directory structure should be modified to match:
Click for details...
```
HAC
├── human
| ├── videos
| | ├── ...
| ├── flow
| | ├── ...
| ├── audio
| | ├── ...
├── animal
| ├── videos
| | ├── ...
| ├── flow
| | ├── ...
| ├── audio
| | ├── ...
├── cartoon
| ├── videos
| | ├── ...
| ├── flow
| | ├── ...
| ├── audio
| | ├── ...
```
Download the pretrained weights similar to EPIC-Kitchens Dataset and put under the HAC-rgb-flow-audio/pretrained_models
directory.
Video and Audio
Click for details...
```
cd HAC-rgb-flow-audio
```
```
python train_video_flow_audio_HAC_SimMMDG.py --use_video --use_audio -s 'animal' 'cartoon' -t 'human' --lr 1e-4 --bsz 16 --nepochs 10 --datapath /path/to/HAC/
```
```
python train_video_flow_audio_HAC_SimMMDG.py --use_video --use_audio -s 'human' 'cartoon' -t 'animal' --lr 1e-4 --bsz 16 --nepochs 10 --datapath /path/to/HAC/
```
```
python train_video_flow_audio_HAC_SimMMDG.py --use_video --use_audio -s 'human' 'animal' -t 'cartoon' --lr 1e-4 --bsz 16 --nepochs 10 --datapath /path/to/HAC/
```
Video and Flow
Click for details...
```
cd HAC-rgb-flow-audio
```
```
python train_video_flow_audio_HAC_SimMMDG.py --use_video --use_flow -s 'animal' 'cartoon' -t 'human' --lr 1e-4 --bsz 16 --nepochs 20 --datapath /path/to/HAC/
```
```
python train_video_flow_audio_HAC_SimMMDG.py --use_video --use_flow -s 'human' 'cartoon' -t 'animal' --lr 1e-4 --bsz 16 --nepochs 20 --datapath /path/to/HAC/
```
```
python train_video_flow_audio_HAC_SimMMDG.py --use_video --use_flow -s 'human' 'animal' -t 'cartoon' --lr 1e-4 --bsz 16 --nepochs 20 --datapath /path/to/HAC/
```
Flow and Audio
Click for details...
```
cd HAC-rgb-flow-audio
```
```
python train_video_flow_audio_HAC_SimMMDG.py --use_flow --use_audio -s 'animal' 'cartoon' -t 'human' --lr 1e-4 --bsz 16 --nepochs 15 --datapath /path/to/HAC/
```
```
python train_video_flow_audio_HAC_SimMMDG.py --use_flow --use_audio -s 'human' 'cartoon' -t 'animal' --lr 1e-4 --bsz 16 --nepochs 15 --datapath /path/to/HAC/
```
```
python train_video_flow_audio_HAC_SimMMDG.py --use_flow --use_audio -s 'human' 'animal' -t 'cartoon' --lr 1e-4 --bsz 16 --nepochs 20 --datapath /path/to/HAC/
```
Video and Flow and Audio
Click for details...
```
cd HAC-rgb-flow-audio
```
```
python train_video_flow_audio_HAC_SimMMDG.py --use_video --use_flow --use_audio -s 'animal' 'cartoon' -t 'human' --lr 1e-4 --bsz 16 --nepochs 15 --datapath /path/to/HAC/
```
```
python train_video_flow_audio_HAC_SimMMDG.py --use_video --use_flow --use_audio -s 'human' 'cartoon' -t 'animal' --lr 1e-4 --bsz 16 --nepochs 10 --datapath /path/to/HAC/
```
```
python train_video_flow_audio_HAC_SimMMDG.py --use_video --use_flow --use_audio -s 'human' 'animal' -t 'cartoon' --lr 1e-4 --bsz 16 --nepochs 15 --datapath /path/to/HAC/
```
Contact
If you have any questions, please send an email to donghaospurs@gmail.com
Citation
If you find our work useful in your research please consider citing our paper:
@inproceedings{dong2023SimMMDG,
title={Sim{MMDG}: A Simple and Effective Framework for Multi-modal Domain Generalization},
author={Dong, Hao and Nejjar, Ismail and Sun, Han and Chatzi, Eleni and Fink, Olga},
booktitle={Advances in Neural Information Processing Systems (NeurIPS)},
year={2023}
}
Related Projects
MOOSA: Towards Multimodal Open-Set Domain Generalization and Adaptation through Self-supervision
MultiOOD: Scaling Out-of-Distribution Detection for Multiple Modalities
Acknowledgement
Many thanks to the excellent open-source projects DomainAdaptation.