SuperClass: A Simple Classification Approach to Vision-Language Pre-Training
SuperClass, a novel classification-based method for vision-language pre-training, achieves comparable performance to contrastive approaches like CLIP, while being simpler, more efficient, and demonstrating strong scaling capabilities.
CMAL: Enhancing Vision-Language Pre-Training with Cross-Modal Associative Learning
CMAL, a novel cross-modal associative learning framework, enhances vision-language pre-training by addressing the limitations of traditional contrastive learning methods through anchor point detection, cross-modal associative prompting, and associative mapping classification.
대비 지역화 언어-이미지 사전 훈련 (CLOC): CLIP의 지역화 기능 향상
CLOC는 CLIP의 이미지-텍스트 정렬 기능을 지역 수준까지 향상시켜, 이미지 내 특정 영역에 대한 이해도를 높인 새로운 사전 훈련 프레임워크입니다.
対照的な局所化言語-画像事前学習：CLIPの地域レベルの理解を強化
本稿では、画像全体とテキストの整合性に加えて、画像の特定領域とテキストの整合性を学習することで、CLIPの地域レベルの理解能力を向上させる新しい事前学習フレームワーク「CLOC」を提案する。
Contrastive Localized Language-Image Pre-Training (CLOC): Enhancing CLIP for Fine-Grained Visual Understanding in Multimodal Large Language Models
CLOC, a novel pre-training framework, enhances CLIP's localization capabilities by incorporating region-text alignment, enabling improved performance in tasks requiring fine-grained visual understanding, particularly within Multimodal Large Language Models (MLLMs).
© 2024 by Linnk AI