21 datasets found

MiniGPT-v2 is a vision-language model that uses a unified interface for multi-task learning.

Perceptual grouping in contrastive vision-language models.

InternLM-XC is a vision-language large model that supports images with any aspect ratio from 336 pixels up to 4K HD, facilitating its deployment in real-world contexts.

InternLM2 is a vision-language large model that supports images with any aspect ratio from 336 pixels up to 4K HD, facilitating its deployment in real-world contexts.

InternLM-XComposer2 is a vision-language large model that supports images with any aspect ratio from 336 pixels up to 4K HD, facilitating its deployment in real-world contexts.

InternLM-XComposer2-4KHD is a vision-language large model that supports images with any aspect ratio from 336 pixels up to 4K HD, facilitating its deployment in real-world...

RH20T-P is a primitive-level robotic dataset built upon RH20T, containing about 33000 video clips covering 44 diverse and complicated robotic tasks.

UniDiff is a unified vision-language model that integrates discriminative and generative capabilities in vision-language tasks.

DataComp-10M is used as a pretraining dataset

CC3M and CC12M are used as datasets for training and evaluation

Self-supervised contrastive learning models, such as CLIP, have set new benchmarks for vision-language models in many downstream tasks. However, their dependency on rigid...

When and why Vision-Language Models behave like Bags-of-Words, and what to do about it?

The dataset used in the paper is Conceptual Captions 12M (CC12M) and RedCaps.

The dataset used in the paper is Conceptual Captions 3M (CC3M), Conceptual Captions 12M (CC12M), RedCaps, and LAION-400M.

A method for learning to prompt for vision-language models.

Open-CLIP dataset contains pre-trained CLIP models.

Conceptual Captions (CC-3M) is a large-scale dataset of 300,000 image-caption pairs.

CC-3M is a large-scale dataset of 300,000 image-caption pairs.

CC-Neg is a large-scale dataset of 228, 246 image-caption pairs along with high quality negated captions as distractors.

The dataset used in this paper is the Playhouse and AndroidEnv environments.