New BS-cRT Method Boosts Long-Tailed Recognition Accuracy

Researchers have developed a new baseline method called BS-cRT for long-tailed recognition tasks, which aims to improve accuracy by retraining only the classifier after initial training. This two-stage procedure involves training a backbone and cosine classifier with Balanced Softmax, then freezing the backbone and updating only the classifier on balanced episodic batches. The BS-cRT method consistently enhances few-shot accuracy across various datasets like CIFAR-100-LT, CIFAR-10-LT, ImageNet-LT, and Places-LT, showing significant gains, particularly at higher imbalance factors. AI

IMPACT This research offers a practical baseline for improving AI's ability to recognize rare classes in datasets, potentially enhancing performance in real-world applications with imbalanced data.

RANK_REASON The cluster contains an academic paper detailing a new method for a machine learning task. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Balanced Softmax
• BS-cRT
• CIFAR-100-LT
• CIFAR-10-LT
• Counterfactual Boundary Risk Minimization
• ImageNet-LT
• Places-LT

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →