MoA Lee 2025 WACV — Independent Reproduction
This repository contains model checkpoints from an independent reproduction of:
Domain Generalization using Large Pretrained Models with Mixture-of-Adapters
Gyuseong Lee, Wooseok Jang, Jinhyeon Kim, Jaewoo Jung, and Seungryong Kim
IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2025
This repository is not affiliated with or maintained by the original authors.
Model description
The checkpoints use an OpenCLIP ViT-B/16 backbone pretrained on LAION-2B:
- Base model: laion/CLIP-ViT-B-16-laion2B-s34B-b88K
- Dataset: PACS
- Task: Domain generalization
- Framework: PyTorch
Four leave-one-domain-out checkpoints are provided. Each checkpoint corresponds to one held-out PACS target domain.
Checkpoints
| Held-out target domain | Checkpoint |
|---|---|
| Art painting | pacs/art_painting.pt |
| Cartoon | pacs/cartoon.pt |
| Photo | pacs/photo.pt |
| Sketch | pacs/sketch.pt |
Checkpoint format
Each .pt checkpoint contains:
- model state
- optimizer state
- training step
- evaluation metrics
- model metadata
The checkpoints require the accompanying Engram MoA implementation for model construction and evaluation.
License and attribution
The independently implemented Engram MoA code and checkpoints are released under the MIT License.
The checkpoints are based on OpenCLIP ViT-B/16 pretrained weights. OpenCLIP and the corresponding base model remain subject to their original license and attribution requirements.
Citation
@inproceedings{DBLP:conf/wacv/LeeJKJK25,
author = {Gyuseong Lee and
Wooseok Jang and
Jinhyeon Kim and
Jaewoo Jung and
Seungryong Kim},
title = {Domain Generalization using Large Pretrained Models with Mixture-of-Adapters},
booktitle = {{IEEE/CVF} Winter Conference on Applications of Computer Vision, {WACV}
2025, Tucson, AZ, USA, February 26 - March 6, 2025},
pages = {8259--8269},
publisher = {{IEEE}},
year = {2025},
url = {https://doi.org/10.1109/WACV61041.2025.00801},
doi = {10.1109/WACV61041.2025.00801},
timestamp = {Wed, 23 Apr 2025 11:40:59 +0200},
biburl = {https://dblp.org/rec/conf/wacv/LeeJKJK25.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}