# High-precision image understanding

A vision-language model based on the OpenCLIP-ViT-B-16 architecture. By introducing test-time registers to optimize the internal representation, it solves the problem of feature map artifacts.

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

CLIP ConvNeXt-XXLarge model trained on LAION-2B dataset, implemented with OpenCLIP framework, the first non-ViT architecture achieving >79% ImageNet zero-shot accuracy

CLIP ConvNeXt-XXLarge model trained on LAION-2B dataset using OpenCLIP framework, the first non-ViT image tower CLIP model achieving >79% ImageNet top-1 zero-shot accuracy