Uniquery improves pedestrian attribute recognition across datasets
UniQuery: Unifying Vision Models with a Joint Query Head for Cross-Domain Pedestrian Attribute Recognition
Computer Vision and Pattern Recognition
Summary
Pedestrian Attribute Recognition helps identify features like gender and clothing from images, useful for finding people in videos. The authors found it's hard to recognize detailed features when pictures vary a lot between cameras or settings. They created UniQuery, a method that combines two image analysis models in a clever way so they work better together than separately. Testing showed UniQuery more accurately identifies these features on several benchmark datasets, especially when recognizing people from new camera setups.
What this means in practice
- •For surveillance system engineers: Improve accuracy in recognizing person attributes across different camera views by using unified model fusion as introduced by UniQuery.
- •For security operations teams: Enhance identity retrieval from video footage by deploying attribute recognition models trained with joint query fusion for better cross-camera consistency.
Authors
Nguyen Dang Duy Tan, Linh T. T. Tran, Huynh Nguyen Phat, Vo Thanh Nguyen
Abstract
Pedestrian Attribute Recognition (PAR) predicts semantic attributes of individuals, such as gender, age, clothing and accessories, and its descriptions support attribute-based person retrieval and person re-identification. However, accurately recognizing fine-grained attributes across domains remains challenging due to occlusion, appearance variations, and shifts in image distributions. Ensembles of independently trained foundation encoders combine their logits with fixed fusion weights, so every attribute uses the same mix and the encoders are never trained together. We propose UniQuery, which instead jointly fine-tunes DINOv3 and CLIP for attribute-specific fusion. It combines two query heads, one for each encoder, with a joint query head whose attribute queries attend to tokens from both encoders, and the prediction is the average of the logits of the three query heads. We evaluate UniQuery on the benchmark of the UPAR Challenge 2027 at the WACV 2027 Real-World Surveillance workshop, where models are trained on PA100k, PETA and Market-1501 with a shared set of attributes. In leave-one-dataset-out experiments, UniQuery outperforms the independent ensemble in Challenge_Avg on both held-out datasets. Ablations show that learning fusion over independently trained encoders does not improve upon logit averaging in the mean over the two held-out datasets, while freezing the encoders lowers the Challenge_Avg of UniQuery. On the hidden test set of the benchmark, UniQuery achieves a Challenge_Avg of 0.8001.