🐾 AniMo: Species-Aware Model for Text-Driven Animal Motion Generation (CVPR 2025) 🦁

🚀 Getting Started

🛠️ Environment

conda env create -f environment.yml
conda activate animo
pip install git+https://github.com/openai/CLIP.git

📊 Data

Please follow the instructions in ./data_generation/README.md to obtain the AniMo4D dataset. 📂

🏋️‍♂️ Training AniMo Model

You may also need to download evaluation models in ./models/text_mot_match/README.md and glove files in ./glove/README.md to run the scripts. ⚙️

🔢 Train RVQ

python train_vq.py \
  --name rvq \
  --gpu_id 0 \
  --batch_size 128 \
  --num_quantizers 6 \
  --max_epoch 50 \
  --quantize_dropout_prob 0.2 \
  --gamma 0.05 \
  --warm_up_iter 80000 \
  --checkpoints_dir ckpt/animo

🎭 Train Base Transformer

python train_t2m_transformer.py \
  --name mtrans \
  --gpu_id 0 \
  --batch_size 64 \
  --vq_name rvq \
  --max_epoch 80 \
  --checkpoints_dir ckpt/animo

➕ Train Residual Transformer

python train_res_transformer.py \
  --name rtrans \
  --gpu_id 0 \
  --batch_size 64 \
  --vq_name rvq \
  --cond_drop_prob 0.2 \
  --max_epoch 80 \
  --share_weight \
  --checkpoints_dir ckpt/animo

📈 Evaluation

🔍 Evaluate RVQ Reconstruction:

python eval_t2m_vq.py \
  --gpu_id 0 \
  --name rvq \
  --ext eval_reconstruction \
  --checkpoints_dir ckpt/animo

✨ Evaluate Text2motion Generation:

python eval_t2m_trans_res.py \
  --res_name rtrans \
  --name mtrans \
  --vq_name rvq \
  --gpu_id 0 \
  --cond_scale 4 \
  --time_steps 10 \
  --ext eval_generation \
  --which_epoch latest \
  --checkpoints_dir ckpt/animo

📚Citation

@inproceedings{wang2025animo,
  title={AniMo: Species-Aware Model for Text-Driven Animal Motion Generation},
  author={Wang, Xuan and Ruan, Kai and Zhang, Xing and Wang, Gaoang},
  booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
  pages={1929--1939},
  year={2025}
}

💖 Acknowlegements

We sincerely thank the open-sourcing of these works where our code is based on: MoMask, Text-to-motion, AttT2M, HumanML3D, OmniMotionGPT and T2M-GPT.