Advances in Mitigating Bias and Improving Generalization in Computer Vision

The field of computer vision is moving towards addressing the long-standing issues of bias and generalization in various tasks such as object detection, image classification, and segmentation. Researchers are proposing innovative solutions to mitigate the effects of imbalanced datasets and improve the performance of models on rare or unseen categories. Notable advancements include the development of adaptive distillation methods, unbiased recovery and relabeling techniques, and novel prototype enhancement strategies. These approaches have shown significant improvements in state-of-the-art performance and have the potential to enable more robust and generalizable computer vision systems. Noteworthy papers include Rectifying Soft-Label Entangled Bias in Long-Tailed Dataset Distillation, which proposes an Adaptive Soft-label Alignment module to calibrate entangled biases, and Unleashing the Power of Vision-Language Models for Long-Tailed Multi-Label Visual Recognition, which introduces a correlation adaptation prompt network to model label correlations from CLIP's textual encoder.

Sources

Rectifying Soft-Label Entangled Bias in Long-Tailed Dataset Distillation

State and Scene Enhanced Prototypes for Weakly Supervised Open-Vocabulary Object Detection

VK-Det: Visual Knowledge Guided Prototype Learning for Open-Vocabulary Aerial Object Detection

PromptMoE: Generalizable Zero-Shot Anomaly Detection via Visually-Guided Prompt Mixtures

Exploring Weak-to-Strong Generalization for CLIP-based Classification

Mitigating Long-Tail Bias in HOI Detection via Adaptive Diversity Cache

Rethinking Long-tailed Dataset Distillation: A Uni-Level Framework with Unbiased Recovery and Relabeling

DualGazeNet: A Biologically Inspired Dual-Gaze Query Network for Salient Object Detection

Unleashing the Power of Vision-Language Models for Long-Tailed Multi-Label Visual Recognition

RefOnce: Distilling References into a Prototype Memory for Referring Camouflaged Object Detection

OVOD-Agent: A Markov-Bandit Framework for Proactive Visual Reasoning and Self-Evolving Detection

Built with on top of