MuCo: Multi-turn Contrastive Learning for Multimodal Embedding Model (CVPR 2026)

Welcom to the official Pytorch implementation of MuCo!

Authors:

Geonmo Gu^1,3, Byeongho Heo¹, Jaemyung Yu¹, Jaehui Hwang¹, Taekyung Kim¹, Sangmin Lee³, HeeJae Jun², Yoohoon Kang², Sangdoo Yun*¹, Dongyoon Han*¹

¹ NAVER AI Lab ² NAVER AI Search Platform ³ Korea University

* Corresponding authors.

🚀 News

• April 16, 2026 - Models and evaluation code are released!
• February 21, 2026 - MuCo is accepted to CVPR 2026!

📚 M3T dataset

[🤗 naver-ai/M3T]

📂 MuCo Models

Model	Avg. MMEB
[🤗 MuCo-2B]	70.1
[🤗 MuCo-7B]	74.2

Note: Performance has been further optimized during the code release preparation. 😊

---

🛠️ Installation

$ pip install -r requirements.txt

🗂️ Dataset Preparation

$ DATASET_FOLDER="./dataset" # Set your path here
$ hf download naver-ai/M3T --repo-type dataset --local-dir ${DATASET_FOLDER}
$ cd ${DATASET_FOLDER}
$ python download_M3T_images.py
$ python download_and_unzip_MMEB_images.py

🔥 Training

Scripts will be updated soon.

💯 Evaluation

torchrun --nproc_per_node=8 --master_port=10000 eval_mmeb.py \
    --pooling eos \
    --normalize \
    --per_device_eval_batch_size 64 \
    --model_name naver-ai/MuCo-2B \
    --data_basedir ${DATASET_FOLDER}/MMEB_eval \
    --encode_output_path ./results/MuCo-2B_MMEB

Citation

@inproceedings{gu2026muco,
    title={MuCo: Multi-turn Contrastive Learning for Multimodal Embedding Model},
    author={Gu, Geonmo and Heo, Byeongho and Yu, Jaemyung and Hwang, Jaehui and Kim, Taekyung and Lee, Sangmin and Jun, HeeJae and Kang, Yoohoon and Yun, Sangdoo and Han, Dongyoon},
    year={2026},
    booktitle={Conference on Computer Vision and Pattern Recognition (CVPR)},
}

License

MuCo
Copyright (c) 2026-present NAVER Cloud Corp.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

     http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.