【ICLR 2024 🔥】LanguageBind: Extending Video-Language Pretraining to N-modality by Language-based Semantic Alignment

If you like our project, please give us a star ⭐ on GitHub for latest update.

[![hf_space](https://img.shields.io/badge/🤗-Open%20In%20Spaces-blue.svg)](https://huggingface.co/spaces/LanguageBind/LanguageBind) [![Dataset meta](https://img.shields.io/badge/%F0%9F%A4%97%20Dataset-VIDAL-blue)](https://huggingface.co/datasets/LanguageBind/VIDAL-Depth-Thermal) [![arXiv](https://img.shields.io/badge/Arxiv-2310.01852-b31b1b.svg?logo=arXiv)](https://arxiv.org/abs/2310.01852) [![wechat](https://img.shields.io/badge/量子位%20-black)](https://mp.weixin.qq.com/s/EFqLv_Euf5VU024zOtzkkg) [![jiqizhixin](https://img.shields.io/badge/机器之心%20-black)](https://mp.weixin.qq.com/s/E5Tazm_vz1CADMwV0tdhnw) [![zhihu](https://img.shields.io/badge/知乎-0084FF)](https://zhuanlan.zhihu.com/p/660567767) [![License](https://img.shields.io/badge/Code%20License-MIT-yellow)](https://github.com/PKU-YuanGroup/LanguageBind/blob/main/LICENSE) [![Data License](https://img.shields.io/badge/Dataset%20license-CC--BY--NC%204.0-orange)](https://github.com/PKU-YuanGroup/LanguageBind/blob/main/DATASET_LICENSE) [![Hits](https://hits.seeyoufarm.com/api/count/incr/badge.svg?url=https%3A%2F%2Fgithub.com%2FPKU-YuanGroup%2FLanguageBind&count_bg=%2379C83D&title_bg=%23555555&icon=&icon_color=%23E7E7E7&title=Visitor&edge_flat=false)](https://hits.seeyoufarm.com) [![GitHub issues](https://img.shields.io/github/issues/PKU-YuanGroup/LanguageBind?color=critical&label=Issues)](https://github.com/PKU-YuanGroup/LanguageBind/issues?q=is%3Aopen+is%3Aissue) [![GitHub closed issues](https://img.shields.io/github/issues-closed/PKU-YuanGroup/LanguageBind?color=success&label=Issues)](https://github.com/PKU-YuanGroup/LanguageBind/issues?q=is%3Aissue+is%3Aclosed)

[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/languagebind-extending-video-language/zero-shot-audio-classification-on-audioset)](https://paperswithcode.com/sota/zero-shot-audio-classification-on-audioset?p=languagebind-extending-video-language)
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/languagebind-extending-video-language/zero-shot-audio-classification-on-vgg-sound)](https://paperswithcode.com/sota/zero-shot-audio-classification-on-vgg-sound?p=languagebind-extending-video-language)
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/languagebind-extending-video-language/zero-shot-text-to-audio-retrieval-on-clotho)](https://paperswithcode.com/sota/zero-shot-text-to-audio-retrieval-on-clotho?p=languagebind-extending-video-language)
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/languagebind-extending-video-language/zero-shot-scene-classification-unified)](https://paperswithcode.com/sota/zero-shot-scene-classification-unified?p=languagebind-extending-video-language)
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/languagebind-extending-video-language/zero-shot-classification-unified-classes-on)](https://paperswithcode.com/sota/zero-shot-classification-unified-classes-on?p=languagebind-extending-video-language)
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/languagebind-extending-video-language/zero-shot-video-retrieval-on-msvd)](https://paperswithcode.com/sota/zero-shot-video-retrieval-on-msvd?p=languagebind-extending-video-language)
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/languagebind-extending-video-language/zero-shot-environment-sound-classification-on-1)](https://paperswithcode.com/sota/zero-shot-environment-sound-classification-on-1?p=languagebind-extending-video-language)
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/languagebind-extending-video-language/zero-shot-text-to-audio-retrieval-on)](https://paperswithcode.com/sota/zero-shot-text-to-audio-retrieval-on?p=languagebind-extending-video-language)
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/languagebind-extending-video-language/zero-shot-video-retrieval-on-activitynet)](https://paperswithcode.com/sota/zero-shot-video-retrieval-on-activitynet?p=languagebind-extending-video-language)
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/languagebind-extending-video-language/zero-shot-video-retrieval-on-msr-vtt)](https://paperswithcode.com/sota/zero-shot-video-retrieval-on-msr-vtt?p=languagebind-extending-video-language)
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/languagebind-extending-video-language/zero-shot-video-retrieval-on-didemo)](https://paperswithcode.com/sota/zero-shot-video-retrieval-on-didemo?p=languagebind-extending-video-language)
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/languagebind-extending-video-language/zero-shot-action-recognition-on-kinetics)](https://paperswithcode.com/sota/zero-shot-action-recognition-on-kinetics?p=languagebind-extending-video-language)

💡 I also have other vision-language projects that may interest you ✨.

> [**Video-LLaVA: Learning United Visual Representation by Alignment Before Projection**](https://arxiv.org/abs/2311.10122)
> Bin Lin, Yang Ye, Bin Zhu, Jiaxi Cui, Munan Ning, Peng Jin, Li Yuan
[![github](https://img.shields.io/badge/-Github-black?logo=github)](https://github.com/PKU-YuanGroup/Video-LLaVA) [![github](https://img.shields.io/github/stars/PKU-YuanGroup/Video-LLaVA.svg?style=social)](https://github.com/PKU-YuanGroup/Video-LLaVA) [![arXiv](https://img.shields.io/badge/Arxiv-2311.10122-b31b1b.svg?logo=arXiv)](https://arxiv.org/abs/2311.10122)
> [**MoE-LLaVA: Mixture of Experts for Large Vision-Language Models**](https://github.com/PKU-YuanGroup/MoE-LLaVA/blob/main/MoE-LLaVA.pdf)
> Bin Lin, Zhenyu Tang, Yang Ye, Jiaxi Cui, Bin Zhu, Peng Jin, Junwu Zhang, Munan Ning, Li Yuan
[![github](https://img.shields.io/badge/-Github-black?logo=github)](https://github.com/PKU-YuanGroup/MoE-LLaVA) [![github](https://img.shields.io/github/stars/PKU-YuanGroup/MoE-LLaVA.svg?style=social)](https://github.com/PKU-YuanGroup/MoE-LLaVA) [![arXiv](https://img.shields.io/badge/Arxiv-2401.15947-b31b1b.svg?logo=arXiv)](https://arxiv.org/abs/2401.15947)
> [**Video-Bench: A Comprehensive Benchmark and Toolkit for Evaluating Video-based Large Language Models**](https://arxiv.org/abs/2311.08046)
> Munan Ning, Bin Zhu, Yujia Xie, Bin Lin, Jiaxi Cui, Lu Yuan, Dongdong Chen, Li Yuan
[![github](https://img.shields.io/badge/-Github-black?logo=github)](https://github.com/PKU-YuanGroup/Video-Bench) [![github](https://img.shields.io/github/stars/PKU-YuanGroup/Video-Bench.svg?style=social)](https://github.com/PKU-YuanGroup/Video-Bench) [![arXiv](https://img.shields.io/badge/Arxiv-2311.16103-b31b1b.svg?logo=arXiv)](https://arxiv.org/abs/2311.16103)

## 📰 News * **[2024.01.27]** 👀👀👀 Our [MoE-LLaVA](https://github.com/PKU-YuanGroup/MoE-LLaVA) is released! A sparse model with 3B parameters outperformed the dense model with 7B parameters. * **[2024.01.16]** 🔥🔥🔥 Our LanguageBind has been accepted at ICLR 2024! We earn the score of 6(3)8(6)6(6)6(6) [here](https://openreview.net/forum?id=QmZKc7UZCy¬eId=OgsxQxAleA). * **[2023.12.15]** 💪💪💪 We expand the 💥💥💥 VIDAL dataset and now have **10M video-text data**. We launch **LanguageBind_Video 1.5**, checking our [model zoo](#-model-zoo). * **[2023.12.10]** We expand the 💥💥💥 VIDAL dataset and now have **10M depth and 10M thermal data**. We are in the process of uploading thermal and depth data on [Hugging Face](https://huggingface.co/datasets/LanguageBind/VIDAL-Depth-Thermal) and expect the whole process to last 1-2 months. * **[2023.11.27]** 🔥🔥🔥 We have updated our [paper](https://arxiv.org/abs/2310.01852) with emergency zero-shot results., checking our ✨ [results](#emergency-results). * **[2023.11.26]** 💥💥💥 We have open-sourced all textual sources and corresponding YouTube IDs [here](DATASETS.md). * **[2023.11.26]** 📣📣📣 We have open-sourced fully fine-tuned **Video & Audio**, achieving improved performance once again, checking our [model zoo](#-model-zoo). * **[2023.11.22]** We are about to release a fully fine-tuned version, and the **HUGE** version is currently undergoing training. * **[2023.11.21]** 💥 We are releasing sample data in [DATASETS.md](DATASETS.md) so that individuals who are interested can further modify the code to train it on their own data. * **[2023.11.20]** 🚀🚀🚀 [Video-LLaVA](https://github.com/PKU-YuanGroup/Video-LLaVA) builds a large visual-language model to achieve 🎉SOTA performances based on LanguageBind encoders. * **[2023.10.23]** 🎶 LanguageBind-Audio achieves 🎉🎉🎉**state-of-the-art (SOTA) performance on 5 datasets**, checking our ✨ [results](#multiple-modalities)! * **[2023.10.14]** 😱 Released a stronger LanguageBind-Video, checking our ✨ [results](#video-language)! The video checkpoint **have updated** on Huggingface Model Hub! * **[2023.10.10]** We provide sample data, which can be found in [assets](assets), and [emergency zero-shot usage](#emergency-zero-shot) is described. * **[2023.10.07]** The checkpoints are available on 🤗 [Huggingface Model](https://huggingface.co/LanguageBind). * **[2023.10.04]** Code and [demo](https://huggingface.co/spaces/LanguageBind/LanguageBind) are available now! Welcome to **watch** 👀 this repository for the latest updates. ## 😮 Highlights ### 💡 High performance, but NO intermediate modality required LanguageBind is a **language-centric** multimodal pretraining approach, **taking the language as the bind across different modalities** because the language modality is well-explored and contains rich semantics. * The following first figure shows the architecture of LanguageBind. LanguageBind can be easily extended to segmentation, detection tasks, and potentially to unlimited modalities. ### ⚡️ A multimodal, fully aligned and voluminous dataset We propose **VIDAL-10M**, **10 Million data** with **V**ideo, **I**nfrared, **D**epth, **A**udio and their corresponding **L**anguage, which greatly expands the data beyond visual modalities. * The second figure shows our proposed VIDAL-10M dataset, which includes five modalities: video, infrared, depth, audio, and language. ### 🔥 Multi-view enhanced description for training We make multi-view enhancements to language. We produce multi-view description that combines **meta-data**, **spatial**, and **temporal** to greatly enhance the semantic information of the language. In addition we further **enhance the language with ChatGPT** to create a good semantic space for each modality aligned language.

## 🤗 Demo * **Local demo.** Highly recommend trying out our web demo, which incorporates all features currently supported by LanguageBind. ```bash python gradio_app.py ``` * **Online demo.** We provide the [online demo](https://huggingface.co/spaces/LanguageBind/LanguageBind) in Huggingface Spaces. In this demo, you can calculate the similarity of modalities to language, such as audio-to-language, video-to-language, and depth-to-image.

## 🚀 Main Results ### Video-Language LanguageBind achieves **state-of-the-art (SOTA) performance on four datasets**, * donates the results of full tuning.

### Multiple Modalities Video-Language, Infrared-Language, Depth-Language, and Audio-Language zero-shot classification, * donates the results of full tuning.

We report text-to-audio results for retrieval, * donates the results of full tuning.

### Emergency results

## 🛠️ Requirements and Installation * Python >= 3.8 * Pytorch >= 1.13.1 * CUDA Version >= 11.6 * Install required packages: ```bash git clone https://github.com/PKU-YuanGroup/LanguageBind cd LanguageBind pip install torch==1.13.1+cu116 torchvision==0.14.1+cu116 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu116 pip install -r requirements.txt ``` ## 🐳 Model Zoo The names in the table represent different encoder models. For example, `LanguageBind/LanguageBind_Video_FT` represents the fully fine-tuned version, while `LanguageBind/LanguageBind_Video` represents the LoRA-tuned version. You can freely replace them in the recommended [API usage](#-api). We recommend using the fully fine-tuned version, as it offers stronger performance.

Modality	LoRA tuning	Fine-tuning
Video	LanguageBind_Video	LanguageBind_Video_FT
Audio	LanguageBind_Audio	LanguageBind_Audio_FT
Depth	LanguageBind_Depth	-
Thermal	LanguageBind_Thermal	-

Version	Tuning	Model size	Num_frames	HF Link	MSR-VTT	DiDeMo	ActivityNet	MSVD
LanguageBind_Video	LoRA	Large	8	Link	42.6	37.8	35.1	52.2
LanguageBind_Video_FT	Full-tuning	Large	8	Link	42.7	38.1	36.9	53.5
LanguageBind_Video_V1.5_FT	Full-tuning	Large	8	Link	42.8	39.7	38.4	54.1
LanguageBind_Video_V1.5_FT	Full-tuning	Large	12	Coming soon
LanguageBind_Video_Huge_V1.5_FT	Full-tuning	Huge	8	Link	44.8	39.9	41.0	53.7
LanguageBind_Video_Huge_V1.5_FT	Full-tuning	Huge	12	Coming soon

## 🤖 API **We open source all modalities preprocessing code.** If you want to load the model (e.g. ```LanguageBind/LanguageBind_Thermal```) from the model hub on Huggingface or on local, you can use the following code snippets! ### Inference for Multi-modal Binding We have provided some sample datasets in [assets](assets) to quickly see how languagebind works. ```python import torch from languagebind import LanguageBind, to_device, transform_dict, LanguageBindImageTokenizer if __name__ == '__main__': device = 'cuda:0' device = torch.device(device) clip_type = { 'video': 'LanguageBind_Video_FT', # also LanguageBind_Video 'audio': 'LanguageBind_Audio_FT', # also LanguageBind_Audio 'thermal': 'LanguageBind_Thermal', 'image': 'LanguageBind_Image', 'depth': 'LanguageBind_Depth', } model = LanguageBind(clip_type=clip_type, cache_dir='./cache_dir') model = model.to(device) model.eval() pretrained_ckpt = f'lb203/LanguageBind_Image' tokenizer = LanguageBindImageTokenizer.from_pretrained(pretrained_ckpt, cache_dir='./cache_dir/tokenizer_cache_dir') modality_transform = {c: transform_dict[c](model.modality_config[c]) for c in clip_type.keys()} image = ['assets/image/0.jpg', 'assets/image/1.jpg'] audio = ['assets/audio/0.wav', 'assets/audio/1.wav'] video = ['assets/video/0.mp4', 'assets/video/1.mp4'] depth = ['assets/depth/0.png', 'assets/depth/1.png'] thermal = ['assets/thermal/0.jpg', 'assets/thermal/1.jpg'] language = ["Training a parakeet to climb up a ladder.", 'A lion climbing a tree to catch a monkey.'] inputs = { 'image': to_device(modality_transform['image'](image), device), 'video': to_device(modality_transform['video'](video), device), 'audio': to_device(modality_transform['audio'](audio), device), 'depth': to_device(modality_transform['depth'](depth), device), 'thermal': to_device(modality_transform['thermal'](thermal), device), } inputs['language'] = to_device(tokenizer(language, max_length=77, padding='max_length', truncation=True, return_tensors='pt'), device) with torch.no_grad(): embeddings = model(inputs) print("Video x Text: \n", torch.softmax(embeddings['video'] @ embeddings['language'].T, dim=-1).detach().cpu().numpy()) print("Image x Text: \n", torch.softmax(embeddings['image'] @ embeddings['language'].T, dim=-1).detach().cpu().numpy()) print("Depth x Text: \n", torch.softmax(embeddings['depth'] @ embeddings['language'].T, dim=-1).detach().cpu().numpy()) print("Audio x Text: \n", torch.softmax(embeddings['audio'] @ embeddings['language'].T, dim=-1).detach().cpu().numpy()) print("Thermal x Text: \n", torch.softmax(embeddings['thermal'] @ embeddings['language'].T, dim=-1).detach().cpu().numpy()) ``` Then returns the following result. ```bash Video x Text: [[9.9989331e-01 1.0667283e-04] [1.3255903e-03 9.9867439e-01]] Image x Text: [[9.9990666e-01 9.3292067e-05] [4.6132666e-08 1.0000000e+00]] Depth x Text: [[0.9954276 0.00457235] [0.12042473 0.8795753 ]] Audio x Text: [[0.97634876 0.02365119] [0.02917843 0.97082156]] Thermal x Text: [[0.9482511 0.0517489 ] [0.48746133 0.5125386 ]] ``` ### Emergency zero-shot Since languagebind binds each modality together, we also found the **emergency zero-shot**. It's very simple to use. ```python print("Video x Audio: \n", torch.softmax(embeddings['video'] @ embeddings['audio'].T, dim=-1).detach().cpu().numpy()) print("Image x Depth: \n", torch.softmax(embeddings['image'] @ embeddings['depth'].T, dim=-1).detach().cpu().numpy()) print("Image x Thermal: \n", torch.softmax(embeddings['image'] @ embeddings['thermal'].T, dim=-1).detach().cpu().numpy()) ``` Then, you will get: ``` Video x Audio: [[1.0000000e+00 0.0000000e+00] [3.1150486e-32 1.0000000e+00]] Image x Depth: [[1. 0.] [0. 1.]] Image x Thermal: [[1. 0.] [0. 1.]] ``` ### Different branches for X-Language task Additionally, LanguageBind can be **disassembled into different branches** to handle different tasks. Note that we do not train Image, which just initialize from OpenCLIP. #### Thermal ```python import torch from languagebind import LanguageBindThermal, LanguageBindThermalTokenizer, LanguageBindThermalProcessor pretrained_ckpt = 'LanguageBind/LanguageBind_Thermal' model = LanguageBindThermal.from_pretrained(pretrained_ckpt, cache_dir='./cache_dir') tokenizer = LanguageBindThermalTokenizer.from_pretrained(pretrained_ckpt, cache_dir='./cache_dir') thermal_process = LanguageBindThermalProcessor(model.config, tokenizer) model.eval() data = thermal_process([r"your/thermal.jpg"], ['your text'], return_tensors='pt') with torch.no_grad(): out = model(**data) print(out.text_embeds @ out.image_embeds.T) ``` #### Depth ```python import torch from languagebind import LanguageBindDepth, LanguageBindDepthTokenizer, LanguageBindDepthProcessor pretrained_ckpt = 'LanguageBind/LanguageBind_Depth' model = LanguageBindDepth.from_pretrained(pretrained_ckpt, cache_dir='./cache_dir') tokenizer = LanguageBindDepthTokenizer.from_pretrained(pretrained_ckpt, cache_dir='./cache_dir') depth_process = LanguageBindDepthProcessor(model.config, tokenizer) model.eval() data = depth_process([r"your/depth.png"], ['your text.'], return_tensors='pt') with torch.no_grad(): out = model(**data) print(out.text_embeds @ out.image_embeds.T) ``` #### Video ```python import torch from languagebind import LanguageBindVideo, LanguageBindVideoTokenizer, LanguageBindVideoProcessor pretrained_ckpt = 'LanguageBind/LanguageBind_Video_FT' # also 'LanguageBind/LanguageBind_Video' model = LanguageBindVideo.from_pretrained(pretrained_ckpt, cache_dir='./cache_dir') tokenizer = LanguageBindVideoTokenizer.from_pretrained(pretrained_ckpt, cache_dir='./cache_dir') video_process = LanguageBindVideoProcessor(model.config, tokenizer) model.eval() data = video_process(["your/video.mp4"], ['your text.'], return_tensors='pt') with torch.no_grad(): out = model(**data) print(out.text_embeds @ out.image_embeds.T) ``` #### Audio ```python import torch from languagebind import LanguageBindAudio, LanguageBindAudioTokenizer, LanguageBindAudioProcessor pretrained_ckpt = 'LanguageBind/LanguageBind_Audio_FT' # also 'LanguageBind/LanguageBind_Audio' model = LanguageBindAudio.from_pretrained(pretrained_ckpt, cache_dir='./cache_dir') tokenizer = LanguageBindAudioTokenizer.from_pretrained(pretrained_ckpt, cache_dir='./cache_dir') audio_process = LanguageBindAudioProcessor(model.config, tokenizer) model.eval() data = audio_process([r"your/audio.wav"], ['your audio.'], return_tensors='pt') with torch.no_grad(): out = model(**data) print(out.text_embeds @ out.image_embeds.T) ``` #### Image Note that our image encoder is the same as OpenCLIP. **Not** as fine-tuned as other modalities. ```python import torch from languagebind import LanguageBindImage, LanguageBindImageTokenizer, LanguageBindImageProcessor pretrained_ckpt = 'LanguageBind/LanguageBind_Image' model = LanguageBindImage.from_pretrained(pretrained_ckpt, cache_dir='./cache_dir') tokenizer = LanguageBindImageTokenizer.from_pretrained(pretrained_ckpt, cache_dir='./cache_dir') image_process = LanguageBindImageProcessor(model.config, tokenizer) model.eval() data = image_process([r"your/image.jpg"], ['your text.'], return_tensors='pt') with torch.no_grad(): out = model(**data) print(out.text_embeds @ out.image_embeds.T) ``` ## 💥 VIDAL-10M The datasets is in [DATASETS.md](DATASETS.md). ## 🗝️ Training & Validating The training & validating instruction is in [TRAIN_AND_VALIDATE.md](TRAIN_AND_VALIDATE.md). ## 👍 Acknowledgement * [OpenCLIP](https://github.com/mlfoundations/open_clip) An open source pretraining framework. * [CLIP4Clip](https://github.com/ArrowLuo/CLIP4Clip) An open source Video-Text retrieval framework. * [sRGB-TIR](https://github.com/rpmsnu/sRGB-TIR) An open source framework to generate infrared (thermal) images. * [GLPN](https://github.com/vinvino02/GLPDepth) An open source framework to generate depth images. ## 🔒 License * The majority of this project is released under the MIT license as found in the [LICENSE](https://github.com/PKU-YuanGroup/LanguageBind/blob/main/LICENSE) file. * The dataset of this project is released under the CC-BY-NC 4.0 license as found in the [DATASET_LICENSE](https://github.com/PKU-YuanGroup/LanguageBind/blob/main/DATASET_LICENSE) file. ## ✏️ Citation If you find our paper and code useful in your research, please consider giving a star :star: and citation :pencil:. ```BibTeX @misc{zhu2023languagebind, title={LanguageBind: Extending Video-Language Pretraining to N-modality by Language-based Semantic Alignment}, author={Bin Zhu and Bin Lin and Munan Ning and Yang Yan and Jiaxi Cui and Wang HongFa and Yatian Pang and Wenhao Jiang and Junwu Zhang and Zongwei Li and Cai Wan Zhang and Zhifeng Li and Wei Liu and Li Yuan}, year={2023}, eprint={2310.01852}, archivePrefix={arXiv}, primaryClass={cs.CV} } ``` ## ✨ Star History [![Star History](https://api.star-history.com/svg?repos=PKU-YuanGroup/LanguageBind&type=Date)](https://star-history.com/#PKU-YuanGroup/LanguageBind&Date) ## 🤝 Contributors