DINOv3 is a series of general visual foundation models developed by Meta AI. It can outperform specialized advanced models in various visual tasks without fine-tuning. The model adopts the Vision Transformer architecture and is pre-trained on 1.689 billion web images. It can generate high-quality dense features and performs excellently in tasks such as image classification, segmentation, and retrieval.

Computer VisionTransformersEnglish

DermLIP is a vision-language model specifically designed for the field of dermatology, trained on the largest dermatology image-text corpus, Derm1M. This model adopts a CLIP-style architecture and can perform various dermatology-related tasks, including zero-shot classification, few-shot learning, cross-modal retrieval, and concept annotation.

MultimodalTransformersEnglish

GME-VARCO-VISION-Embedding is a multimodal embedding model that focuses on calculating the semantic similarity between text, images, and videos in a high-dimensional embedding space, and is particularly good at video retrieval tasks.

MultimodalTransformersEnglish

PHOENIX is a domain-adaptive model based on CLIP/ViT, designed to enhance patent image retrieval capabilities, particularly suitable for retrieving semantically or hierarchically related images rather than exact matches.

Computer VisionSafetensors

ColNomic Embed Multimodal 3B is a 3-billion-parameter multimodal embedding model specifically designed for visual document retrieval tasks, supporting unified encoding of multilingual text and images.

MultimodalSafetensorsMultiple Languages

A multilingual visual retrieval model based on Qwen2.5-VL-3B-Instruct and ColBERT strategy, supporting dynamic input image resolution and multilingual document retrieval.

MultimodalSafetensorsMultiple Languages

A multilingual visual retrieval model based on Qwen2.5-VL-3B-Instruct and ColBERT strategy, supporting dynamic input image resolution and generating ColBERT-style multi-vector text and image representations.

MultimodalTransformersMultiple Languages

A visual retrieval model based on Qwen2-VL-2B-Instruct and ColBERT strategy, capable of generating multi-vector representations for text and images

MultimodalTransformersEnglish

A visual retrieval model based on Qwen2.5-VL-3B-Instruct and ColBERT strategy, capable of generating multi-vector representations for text and images to enable efficient document retrieval.

MultimodalSafetensorsEnglish

This model is trained on paintings from a 2-hour museum tour, aiming to enhance painting feature extraction in image retrieval and zero-shot frame learning.

Computer VisionTransformers

This is a vision-language model based on the OpenCLIP framework, trained on the LAION-2B English subset, excelling in zero-shot image classification and cross-modal retrieval tasks.

MultimodalSafetensors

A vision-language model trained on the LAION-2B English dataset based on the OpenCLIP framework, supporting zero-shot image classification and cross-modal retrieval

MultimodalSafetensors

A vision-language model fine-tuned based on CLIP ViT-L/14, optimized for abstract image-text retrieval tasks

MultimodalTensorboardEnglish

A Japanese CLIP model trained based on the OpenCLIP framework, trained on a dataset of 1.45 billion Japanese image-text pairs, supporting zero-shot image classification and image-text retrieval tasks

MultimodalSafetensorsJapanese

A visual retrieval model based on Qwen2-VL-2B-Instruct and ColBERT strategy, capable of generating multi-vector text and image representations

MultimodalSafetensorsMultiple Languages

A visual retrieval model based on Qwen2-VL-7B-Instruct and ColBERT strategy, supporting multi-vector text and image representation

MultimodalSafetensorsEnglish

InstructCIR is an instruction-aware contrastive learning-based compositional image retrieval model, utilizing ViT-L-224 and Phi-3.5-Mini architectures, focusing on image-text-to-text generation tasks.

MultimodalPytorch

A multimodal model based on the SoViT backbone network, improved with the Sigmoid loss function, supporting zero-shot image classification and image-text retrieval

MultimodalTransformers

VisRAG is a retrieval-augmented generation (RAG) system based on vision-language models (VLM) that can directly embed documents as images, avoiding information loss caused by traditional text parsing.

MultimodalSafetensorsEnglish

SigLIP is an improved multimodal model based on CLIP, employing a superior Sigmoid loss function, pre-trained on the WebLI dataset, and suitable for tasks such as zero-shot image classification and image-text retrieval.

MultimodalTransformers