New fully open source vision encoder OpenVision arrives to improve on OpenAI’s Clip, Google’s SigLIP
1 min read
Summary
OpenVision is a family of AI vision encoders that has been released by the University of California, Santa Cruz (UCSC).
It is intended to be an alternative to Google’s SigLIP and OpenAI’s CLIP and is available under an open Apache 2.0 licence.
A vision encoder is necessary to enable large language models (LLMs) to understand images.
OpenVision comes in 26 variants with between 5.9 million and 632.1 million parameters, allowing developers to choose the most suitable version for the task at hand, and offering optimum performance in server or edge deployments.
OpenVision is particularly suitable for edge deployments in constrained environments such as consumer smartphones, on-site manufacturing cameras and sensors.