Revisiting Vehicle Color Recognition in Long-Tailed Surveillance Scenarios
Researchers have developed a new method to improve vehicle color recognition in surveillance systems, particularly for rare colors. The study utilizes the UFPR-VeSV dataset and employs synthetic data augmentation techniques, including text-conditioned image generation with RunDiffusion/JuggernautXL and image-conditioned color editing with Gemini 2.0 Flash. By combining synthetic data with advanced visual representations and training strategies, the best approach achieved a macro accuracy of 79.7%, an improvement of 8.2 percentage points over previous methods. AI