What AI says about Long-CLIP

Long-CLIP is the official implementation of 'Long-CLIP: Unlocking the Long-Text Capability of CLIP', extending CLIP to handle longer text inputs by increasing the maximum length from 77 to 248 tokens. The repository provides training, evaluation, and inference code, along with checkpoints and demos (including SDXL integration) showing improved long-caption and traditional text-image retrieval. It is designed as a plug-in for any workflow needing long-text support in CLIP-based vision-language tasks and aligns with the ECCV 2024 publication.