The field of computer vision is undergoing significant advancements, with a focus on continual learning and vision-language models. Researchers are exploring ways to improve the performance of these models in various tasks, including image classification, object detection, and visual question answering. One of the key challenges in continual learning is the ability to adapt to new tasks and data distributions without forgetting previous knowledge. To address this, techniques such as synthetic replay, adversarial training, and knowledge distillation are being developed. Noteworthy papers in this area include LoRA-Loop, which proposes a LoRA-enhanced synthetic-replay framework for continual vision-language learning, and Franca, which presents a fully open-source vision foundation model that matches and surpasses the performance of state-of-the-art proprietary models. Additionally, papers like CLIPTTA and HiCroPL are exploring new approaches to test-time adaptation and prompt learning for vision-language models, demonstrating significant improvements in performance and robustness.
Continual Learning and Vision-Language Models
Sources
LoRA-Loop: Closing the Synthetic Replay Cycle for Continual VLM Learning
Franca: Nested Matryoshka Clustering for Scalable Visual Representation Learning
CLIPTTA: Robust Contrastive Vision-Language Test-Time Adaptation
X-Nav: Learning End-to-End Cross-Embodiment Navigation for Mobile Robots
Exploring Scalable Unified Modeling for General Low-Level Vision
Hierarchical Cross-modal Prompt Learning for Vision-Language Models
One Last Attention for Your Vision-Language Model
Visual-Language Model Knowledge Distillation Method for Image Quality Assessment
SeC: Advancing Complex Video Object Segmentation via Progressive Concept Construction
Advancing Visual Large Language Model for Multi-granular Versatile Perception
Quality Text, Robust Vision: The Role of Language in Enhancing Visual Robustness of Vision-Language Models
Cross-Modal Distillation For Widely Differing Modalities
PIG-Nav: Key Insights for Pretrained Image Goal Navigation Models
MaskedCLIP: Bridging the Masked and CLIP Space for Semi-Supervised Medical Vision-Language Pre-training
ViRN: Variational Inference and Distribution Trilateration for Long-Tailed Continual Representation Learning
LMM-Det: Make Large Multimodal Models Excel in Object Detection