InterCMDM

This repository contains materials developed by LY Corporation and is temporarily open-sourced for the purpose of our reseach project.

• Temporary Release: This repository is temporarily available as open-source. Therefore this repository may be turn into read-only or private anytime.
• Attribution: All code and materials in this repository are owned by LY Corporation.

Code of the paper "InterCMDM: Block-Causal Diffusion for Autoregressive Human Interaction Generation" (ECCV 2026). InterCMDM generates autoregressive two-person interactions with a block-causal diffusion model.

uv sync

InterCMDM
├── checkpoints
│   ├── eval_model
│   └── hhi
├── data
│   ├── InterHuman
│   └── Inter-X_Dataset
├── models
├── options
├── utils
├── eval_inter_cmdm.py
├── train_inter_cmdm.py
└── train_tae.py

uv run python train_tae.py \
  --gpu_id 0 \
  --dataset_name interhuman \
  --name tae \
  --batch_size 256 \
  --max_epoch 100

uv run python train_tae.py \
  --gpu_id 0 \
  --dataset_name interx \
  --name tae \
  --batch_size 256 \
  --feature_dim 336 \
  --max_epoch 100

uv run python train_inter_cmdm.py \
  --gpu_id 0 \
  --dataset_name interhuman \
  --name inter_cmdm \
  --vq_name tae \
  --batch_size 64 \
  --max_epoch 500

uv run python train_inter_cmdm.py \
  --gpu_id 0 \
  --dataset_name interx \
  --name inter_cmdm \
  --vq_name tae \
  --batch_size 64 \
  --max_epoch 500

uv run python eval_inter_cmdm.py \
  --gpu_id 0 \
  --dataset_name interhuman \
  --name tae \
  --use_trans false

uv run python eval_inter_cmdm.py \
  --gpu_id 0 \
  --dataset_name interx \
  --name tae \
  --use_trans false

uv run python eval_inter_cmdm.py \
  --gpu_id 0 \
  --dataset_name interhuman \
  --name inter_cmdm \
  --cond_scales 3

uv run python eval_inter_cmdm.py \
  --gpu_id 0 \
  --dataset_name interx \
  --name inter_cmdm \
  --cond_scales 1.5

The data preparation and evaluation setup build on the InterHuman, Inter-X, and InterMask projects. We thank their authors for making their work available.

If you find this project useful, please cite our paper.

@inproceedings{intercmdm2026,
  title     = {InterCMDM: Block-Causal Diffusion for Autoregressive Human Interaction Generation},
  author    = {Yu, Qing and Fujiwara, Kent},
  booktitle = {European Conference on Computer Vision (ECCV)},
  year      = {2026}
}

As this project is temporarily open-sourced, we are not accepting contributions. For feedback or inquiries, please open an issue in this repository.

This code is dedicated to the public domain under CC0 1.0. You may copy, modify, and distribute it without restriction, and the authors make no warranties or guarantees regarding its use.

Additionally, this repository contains third-party software. Refer NOTICE.txt for more details and follow the terms and conditions of their use.