Ex-MCR: Extending Multi-modal Contrastive Representation

Zehan Wang*, Ziang Zhang*, Luping Liu, Yang Zhao, Haifeng Huang, Tao Jin, Zhou Zhao†

Ex-MCR is a training-efficient and paired-data-free method to flexibly learn unified contrastive representation space, by integrating the knowledge of existing MCR spaces.

This implementation provides 3D-image-text-audio unified contrastive representation, obtained by aligning the representation spaces of CLAP (audio-text) and ULIP v2 (3D-image) into the CLIP (image-text).

News

• [10/16/2023] Ex-MCR has been released! [paper] [code]
• [10/11/2023] Source Code of C-MCR has been released! [code]
• [09/22/2023] C-MCR has been accepted by NIPS 2023!🔥🔥🔥 [paper]

File structure

-assets
	[demo samples, including image, audio and 3d model]
-checkpoints
	[pretrained weights for clip,clap,ulip and ex-mcr_clap, ex-mcr_ulip]
-exmcr
	- ULIP [source code of ULIP]
	exmcr_projector.py [the projector of ex-mcr]
	trunks.py [feature extractor of clip, clap and ulip]
	exmcr_model.py [combine projector and trunks together with useful functions]
	type.py
		

Usage

Install enviornments

Install pytorch 1.13+ and other 3rd party dependencies.

conda create -n exmcr python=3.8.16
conda activate exmcr
pip install -r requirements.txt

All feature extractors we use and their pretrained weights are shown below. You need to download the weights for CLAP and ULIP and put them in directory checkpoints and renamed them. The weights for CLIP will be downloaded automatically during the first running.

• CLIP：vit/B-32, you can find the repository here
• CLAP：LAION_AI/CLAP 630k-fusion-best.pt, you can find the repository here and you can download the weight we use here.
• ULIP-2：pointbert, you can find the repository here and you can download the weight we use here

The final structure of checkpoints looks like this:

-checkpoints
	ex_clap.pt
	ex_ulip.pt
	laion_clap_fullset_fusion.pt
	pointbert_ULIP-2.pt

Inference

Extract and compare embeddings in Base-MCR across modalities:

from exmcr.exmcr_model import Ex_MCR
from exmcr.exmcr_model import ModalityType, MCRType
import torch

input = {ModalityType.VISION: ['assets/toilet.jpeg',
                               'assets/BBQ.jpeg',
                               'assets/train.jpeg'],
         ModalityType.TEXT: ['a toilet',
                             'BBQ',
                             'a train'],
         ModalityType.AUDIO:['assets/toilet.wav',
                             'assets/BBQ.wav',
                             'assets/train.wav'],
         ModalityType.PC:['assets/toilet.npy',
                          'assets/BBQ.npy',
                          'assets/train.npy']
         }

device = 'cuda:0' if torch.cuda.is_available() else 'cpu'
model = Ex_MCR(device=device)

# you can get single modality embeddings by using these functions

# v_emb = model.get_vision_embedding(input)
# t_emb = model.get_text_embedding(input)
# a_emb = model.get_audio_embedding(input)
# p_emb = model.get_3d_embedding(input)

embeddings = model.get_embeddings_in_base_mcr(input)

print(
    "Vision x Text:\n",
    torch.softmax(embeddings[ModalityType.VISION] @ embeddings[ModalityType.TEXT].T * 10.0, dim=-1)
)
print(
    "Audio x Text:\n",
    torch.softmax(embeddings[ModalityType.AUDIO] @ embeddings[ModalityType.TEXT].T * 10.0, dim=-1)
)
print(
    "3D x VISION:\n",
    torch.softmax(embeddings[ModalityType.PC] @ embeddings[ModalityType.VISION].T * 10.0, dim=-1)
)
print(
    "Audio x Vision:\n",
    torch.softmax(embeddings[ModalityType.AUDIO] @ embeddings[ModalityType.VISION].T * 10.0, dim=-1)
)
print(
    "3D x Text:\n",
    torch.softmax(embeddings[ModalityType.PC] @ embeddings[ModalityType.TEXT].T * 10.0, dim=-1)
)
print(
    "Audio x 3D:\n",
    torch.softmax(embeddings[ModalityType.AUDIO] @ embeddings[ModalityType.PC].T * 10.0, dim=-1)
)

# Expected output

# Vision x Text:
#  tensor([[0.5737, 0.2242, 0.2021],
#         [0.2454, 0.4756, 0.2790],
#         [0.2541, 0.2530, 0.4929]], device='cuda:3')
# Audio x Text:
#  tensor([[0.9115, 0.0369, 0.0516],
#         [0.1443, 0.7265, 0.1292],
#         [0.1000, 0.0960, 0.8040]], device='cuda:3')
# 3D x VISION:
#  tensor([[0.8635, 0.0850, 0.0515],
#         [0.0761, 0.9125, 0.0114],
#         [0.1239, 0.0781, 0.7980]], device='cuda:3')
# Audio x Vision:
#  tensor([[0.9060, 0.0329, 0.0611],
#         [0.0743, 0.8742, 0.0515],
#         [0.0648, 0.0227, 0.9126]], device='cuda:3')
# 3D x Text:
#  tensor([[0.8052, 0.0909, 0.1039],
#         [0.2447, 0.6035, 0.1518],
#         [0.1757, 0.2094, 0.6149]], device='cuda:3')
# Audio x 3D:
#  tensor([[0.9860, 0.0038, 0.0102],
#         [0.0078, 0.9875, 0.0048],
#         [0.0204, 0.0021, 0.9775]], device='cuda:3')

TODO

• Pre-trained Ex-MCR for ULIP-2, CLIP, CLAP (Unified 3D-image-text-audio representations)
• Code for training
• Code for inference

Citation

If you find this proiect useful in our research, please consider giving a star ⭐ and citation:

@misc{wang2023connecting,
      title={Connecting Multi-modal Contrastive Representations}, 
      author={Zehan Wang and Yang Zhao and Xize Cheng and Haifeng Huang and Jiageng Liu and Li Tang and Linjun Li and Yongqi Wang and Aoxiong Yin and Ziang Zhang and Zhou Zhao},
      year={2023},
      eprint={2305.14381},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}
@misc{wang2023extending,
      title={Extending Multi-modal Contrastive Representations}, 
      author={Zehan Wang and Ziang Zhang and Luping Liu and Yang Zhao and Haifeng Huang and Tao Jin and Zhou Zhao},
      year={2023},
      eprint={2310.08884},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

lf you have any questions or suggestions, feel free to drop us an email ( [email protected], [email protected] ) or open an issue.

Acknowledgement

Thanks to the open source of the following projects: CLIP, CLAP, ULIP, Imagebind.