# Cross-modal Retrieval

UniME is a general embedding learning framework based on multimodal large models, significantly enhancing multimodal embedding capabilities through text discriminative knowledge distillation and hard negative sample-enhanced instruction tuning strategies.

UniME is a general embedding learning model based on a multimodal large model, trained with 336×336 image resolution and ranked first on the MMEB leaderboard.

A multimodal embedding model based on Qwen2.5-Omni-7B, supporting unified embedding representations for cross-lingual text, images, audio, and video

mmE5 is a multimodal multilingual embedding model trained on Llama-3.2-11B-Vision, improving embedding performance through high-quality synthetic data and achieving state-of-the-art results on the MMEB benchmark.

ConceptCLIP is a large-scale vision-language pre-training model enhanced with medical concepts, suitable for various medical imaging modalities, capable of achieving robust performance across multiple medical imaging tasks.

MEXMA-SigLIP is a high-performance CLIP model combining multilingual text encoder and image encoder, supporting 80 languages.

LLM2CLIP is an innovative approach that enhances CLIP's cross-modal capabilities through large language models, significantly improving the discriminative power of visual and text representations.

A multilingual vision-language pre-trained model for the remote sensing field, supporting image-text cross-modal tasks in 10 languages.

A large-scale vision-language model based on Vision Transformer architecture, supporting cross-modal understanding between images and text

High-performance visual embedding model, sharing the same embedding space with nomic-embed-text-v1.5, supporting multimodal applications

High-performance vision embedding model, sharing the same embedding space with nomic-embed-text-v1, supporting multimodal applications

ONNX ported version based on CLIP ViT-B/32 architecture, suitable for image classification and similarity search tasks.

M3D-CLIP is a CLIP model specifically designed for 3D medical imaging, achieving visual and language alignment through contrastive loss.

BLaIR is a language model pre-trained on the Amazon Reviews 2023 dataset, focusing on recommendation and retrieval scenarios, capable of generating powerful product text representations and predicting relevant products.

OWLv2 is a vision-language pre-trained model focused on object detection and localization tasks.

InternVL-14B-224px is a 14B-parameter vision-language foundation model supporting various vision-language tasks.

LanguageBind is a pretrained model that achieves multimodal semantic alignment through language, capable of binding various modalities such as video, audio, depth, and thermal imaging with language to enable cross-modal understanding and retrieval.

Vision-language model based on ViLT architecture, fine-tuned for specific tasks

LanguageBind is a language-centric multimodal pretraining method that achieves semantic alignment by using language as the bridge between different modalities.

LanguageBind is a multimodal model that extends video-language pretraining to N modalities through language-based semantic alignment, accepted by ICLR 2024.

MetaCLIP is an implementation of the CLIP framework applied to CommonCrawl data, aiming to reveal CLIP's training data filtering methods

The MetaCLIP base model is a vision-language model trained on CommonCrawl data for constructing shared image-text embedding spaces.

LanguageBind is a language-centric multimodal pretraining method that uses language as the bond between different modalities to achieve semantic alignment.

LanguageBind is a language-centric multimodal pretraining method that uses language as the bond between different modalities to achieve semantic alignment across video, infrared, depth, audio, and other modalities.

LanguageBind is a multimodal pretraining framework that extends video-language pretraining to N modalities through language semantic alignment, accepted by ICLR 2024.

LanguageBind is a pretraining framework that achieves multimodal semantic alignment through language as the bond, supporting joint learning of various modalities such as video, infrared, depth, and audio with language.

This is a vision-language model based on the CLIP architecture, specifically post-trained on 80 million face images.

A large CLIP model trained on the LAION-2B dataset, using ViT-bigG-14 architecture, supporting cross-modal understanding between images and text

CLIP model developed by OpenAI, based on Vision Transformer architecture, supporting joint understanding of images and text

OpenAI's open-source CLIP model, based on Vision Transformer architecture, supporting cross-modal understanding of images and text

A vision-language model based on the CLIP architecture, supporting zero-shot image classification tasks, trained on the laion dataset

A zero-shot image classification model based on the CLIP architecture, trained using the DataComp dataset, supporting efficient image-text matching tasks.

A vision-language model based on the CLIP architecture, supporting zero-shot image classification tasks

EVA-Giant is a large-scale vision-language model based on the CLIP architecture, supporting zero-shot image classification tasks.

EVA02 is a large-scale vision-language model based on the CLIP architecture, supporting zero-shot image classification tasks.

X-CLIP is an extension of CLIP for general video-language understanding, achieving video classification and video-text retrieval tasks through contrastive learning.

A large-scale vision-language pretrained model based on the Vision Transformer architecture, supporting cross-modal understanding between images and text

A Chinese multimodal pretraining model built on the Oscar framework, initialized with Mengzi-Bert base version, trained on 3.7 million image-text pairs.

A multilingual CLIP text encoder fine-tuned from BERT-base-multilingual, supporting alignment with CLIP visual encoder across 69 languages