# Zero-shot Visual Understanding

The Perception Encoder is a state-of-the-art image and video understanding encoder trained through simple vision-language learning, achieving top performance across various visual tasks.

The Perception Encoder (PE) is a state-of-the-art image and video understanding encoder trained through simple vision-language learning, achieving top performance across various visual tasks.

A large-scale visual encoder model developed by Meta, achieving state-of-the-art performance in various vision tasks through contrastive pre-training and fine-tuning on synthetic video data

AKI is a multimodal foundation model that achieves cross-modal mutual attention (MMA) by unlocking the causal attention mechanism in LLMs, addressing vision-language misalignment without additional parameters or training time.

A specialized version for Visual Question Answering (VQA) fine-tuned based on microsoft/Florence-2-base-ft, capable of interpreting image content and answering related questions

InstructBLIP is the visual instruction-tuned version of BLIP-2, based on the Vicuna-13b language model, designed for vision-language tasks.

A Chinese multimodal chat model trained based on BLIP2, with basic image understanding capabilities, and text dialogue performance consistent with ChatGLM