Code for "HGLTR:Hierarchical knowledge injection for calibrating pre-trained models in long-tail recognition"

In this paper, we propose HGLTR (Hierarchy-Guided Long-Tail Recognition)

Dataset

• Place365: Small images (256 * 256) with easy directory structure - Train and val images. 21G. [link] [hierarchical structure]
• iNaturalist2018: All training and validation images [120GB] [link]
• ImageNet: Training images (Task 1 & 2). 138GB. & Validation images (all tasks). 6.3GB. [link] [details of dataset construction] [hierarchical structure]

Reproduce

• Dependencies: requirements.txt
• Bash script: run.sh

Reference

• openai/CLIP
• shijxcs/LIFT