# Multimodal pre-training

A zero-shot skin disease assessment model based on vision-language pre-training technology, integrating multi-faceted knowledge enhancement to provide an effective tool for skin disease research and diagnosis.

A vision model based on the Vision Transformer architecture, trained using SigLIP (Sigmoid Loss for Language-Image Pretraining), suitable for image understanding tasks.

CoMP-MM-1B is a visual foundation model (VFM) that supports native image resolution input, continuously pre-trained based on SigLIP.

AIM-v2 is an image feature extraction model based on the timm library, designed with large patches for high-resolution image processing.

AIM-v2 is an image encoder model based on the timm library, with a 3B parameter scale, suitable for image feature extraction tasks.

AIM-v2 is an image encoder model based on the timm library, suitable for image feature extraction tasks.

AIM-v2 is an image encoder model developed by Apple, based on a timm-compatible architecture, suitable for image feature extraction tasks.

ResNet101 image encoder based on CLIP framework, extracting image features through Global Average Pooling (GAP)

ResNet50x4 variant model based on the CLIP framework, designed for image feature extraction

A ResNet50 variant based on the visual encoder part of the CLIP model, extracting image features through Global Average Pooling (GAP)

ViT-Huge image encoder based on CLIP architecture, trained on DFN5B dataset, supports quick GELU activation

The visual encoder component of DFN5B-CLIP, based on ViT-Huge architecture, trained with 378x378 resolution images for CLIP model

Vision Transformer model based on CLIP architecture, featuring DFN2B-CLIP image encoder weights released by Apple

Vision model based on SigLIP image encoder with global average pooling, part of the PaliGemma2 model

A ViT-B-16 image encoder model based on SigLIP, using original attention pooling, suitable for image feature extraction tasks.

ViT-Huge visual encoder based on the CLIP framework, trained on the laion2B dataset, supports image feature extraction

Vision Transformer model based on CLIP architecture, specialized in image feature extraction with support for 256x256 resolution input

Vision Transformer model based on CLIP architecture, designed for image feature extraction, trained on the laion2B dataset

Vision Transformer model based on CLIP architecture, designed for image feature extraction, trained using the DataComp XL dataset

A vision Transformer model based on the CLIP architecture, specifically designed for image feature extraction, using ViT-B/16 structure and trained on the DataComp XL dataset

ConvNeXt-XXLarge image encoder based on the CLIP framework, trained by LAION, suitable for multimodal tasks

ConvNeXt Base model based on the CLIP framework, trained on the LAION-Aesthetic dataset, suitable for image feature extraction tasks.

CLIP model based on ViT-Huge architecture, supporting zero-shot image classification tasks

Vision Transformer model trained on LAION-400M dataset, supporting zero-shot image classification tasks

Vision Transformer model trained on LAION-400M dataset, compatible with both OpenCLIP and timm frameworks

CLIP model with ResNet50 architecture trained on the CC12M dataset, supporting zero-shot image classification tasks

ResNet50 model trained on the YFCC-15M dataset, compatible with both open_clip and timm frameworks, supporting zero-shot image classification tasks.

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.

A compact vision-language model based on CLIP architecture, designed for efficient zero-shot image classification

Vision Transformer model based on OpenCLIP framework, fine-tuned for zero-shot image classification tasks

InternViT-6B-448px-V1-2 is a foundational vision model with a feature backbone, comprising 55.4 million parameters, supporting image processing at 448x448 pixels.

SigLIP is a multimodal model pre-trained on the WebLi dataset, employing an improved sigmoid loss function, suitable for zero-shot image classification and image-text retrieval tasks.

SigLIP is a vision-language model pre-trained on the WebLi dataset, employing an improved Sigmoid loss function, excelling in image classification and image-text retrieval tasks.

The ProtST framework enhances the pre-training and understanding of protein sequences through biomedical text, constructs the ProtDescribe dataset, designs three pre-training tasks, and supports supervised learning and zero-shot prediction.

AltCLIP-m18 is a CLIP model supporting 18 languages for image-text matching tasks.

A CLIP ConvNeXt-XXLarge model trained on the LAION-2B dataset, implemented using the OpenCLIP framework, focusing on zero-shot image classification tasks.

Language-independent Layout Transformer (LiLT) provides a LayoutLM-like model for any language by combining pre-trained RoBERTa (English) with a pre-trained language-independent layout transformer (LiLT).