[ICLR 25] The implementation of paper Diff-Prompt: Diffusion-Driven Prompt Generator with Mask Supervision.
Python
52
8 commits
updated Jul 17, 2025
We use AutoencoderKL from diffusers for training Mask-VAE.
cd unique_vae
pip install -r requirements.txt
python run.py --config configs/mask_vae_v2.yaml
Set hyper parameters in the configs/mask_vae_v2.yaml.
We provide pre-trained model here, and a simple demo for using the Mask-VAE in quick_start.py file. Reconstruction results are saved in the asset folder.
python quick_start.py
cd DiT-main
pip install -r requirements.txt
Place 'bert-base-uncased' under DiT-main dir.
torchrun --nnodes=1 --nproc_per_node 4 train.py --global-batch-size 128
Model weights are saved in results dir.
replace ckpt_path,repository_id,HF_TOKEN in this file.
python save_dit_model_safe.py
We provide pre-trained model here.
Follow GLIP to build maskrcnn_benckmark package. Prepare bert-base-uncased under GLIP dir and place dataset under GLIP/DATASET dir.
sh dif_prompt_v3.sh
If you use our work, please consider citing:
@inproceedings{yan2025diff,
title={Diff-Prompt: Diffusion-driven Prompt Generator with Mask Supervision},
author={Yan, Weicai and Lin, Wang and Guo, Zirun and Wang, Ye and Feng, Fangming and Yang, Xiaoda and Wang, Zehan and Jin, Tao},
booktitle={The Thirteenth International Conference on Learning Representations},
year={2025}
}
Our code is based on the following repositories. We thank the authors for releasing their code. If you use our model and code, please consider citing these works as well.
[ICLR 25] The implementation of paper Diff-Prompt: Diffusion-Driven Prompt Generator with Mask Supervision.
Python
52
8 commits
updated Jul 17, 2025
We use AutoencoderKL from diffusers for training Mask-VAE.
cd unique_vae
pip install -r requirements.txt
python run.py --config configs/mask_vae_v2.yaml
Set hyper parameters in the configs/mask_vae_v2.yaml.
We provide pre-trained model here, and a simple demo for using the Mask-VAE in quick_start.py file. Reconstruction results are saved in the asset folder.
python quick_start.py
cd DiT-main
pip install -r requirements.txt
Place 'bert-base-uncased' under DiT-main dir.
torchrun --nnodes=1 --nproc_per_node 4 train.py --global-batch-size 128
Model weights are saved in results dir.
replace ckpt_path,repository_id,HF_TOKEN in this file.
python save_dit_model_safe.py
We provide pre-trained model here.
Follow GLIP to build maskrcnn_benckmark package. Prepare bert-base-uncased under GLIP dir and place dataset under GLIP/DATASET dir.
sh dif_prompt_v3.sh
If you use our work, please consider citing:
@inproceedings{yan2025diff,
title={Diff-Prompt: Diffusion-driven Prompt Generator with Mask Supervision},
author={Yan, Weicai and Lin, Wang and Guo, Zirun and Wang, Ye and Feng, Fangming and Yang, Xiaoda and Wang, Zehan and Jin, Tao},
booktitle={The Thirteenth International Conference on Learning Representations},
year={2025}
}
Our code is based on the following repositories. We thank the authors for releasing their code. If you use our model and code, please consider citing these works as well.