China’s AI-focused company Cloudwalk,Dear Utol (2025): Maniac Photographer Episode 26 known for supplying facial recognition technology to the Bank of China, has made significant progress in the field of large-scale visual models. Its foundational large model for pedestrian analysis has outperformed products developed by renowned universities, companies, and research institutions across four datasets, namely PA-100K, RAP V2, PETA, and HICO-DET, setting new world records. Notably, the model achieved an impressive fine-tuning accuracy of 92.89% on the PA100K dataset. Cloudwalk uses over 2 billion data points, including a vast amount of unlabeled datasets and multimodal data, to develop the foundational model, enabling it to extract robust features suitable for a variety of pedestrian-related tasks. [36Kr, in Chinese]
