Pose-Enriched Text-Image Person Re-Identification Via Cross-Modal Learning
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Text-Image Person Reidentification is an emerging area within the field of Person Reidentification and a crucial research domain in Computer Vision. It has extensive applications in public security surveillance. In Text-Image Person Reidentification, each input RGB image is associated with a textual description of the image. Textual Description often comprise of color, appearance, gender and attributes. Existing works utilize Pose information to align the human body parts, which is vital for handling viewpoint and occlusion challenges. This Paper proposes a novel approach by incorporating the pose cues to the Textual description to form a Pose-Enriched Text. The Confidence score of the keypoints and Scoring Algorithm are used to obtain a person orientation and it is termed as Pose Label. Then, the pose label is merged with the textual caption to create a Pose-Enriched Textual Caption. A visual language model, Contrastive Language-Image Pretraining (CLIP), is used to align Pose-Enriched Text and Image features in a shared embedding space. This aids in matching query text with a gallery of images leading to a significant improvement in TextImage ReID model performance. Our model's performance is evaluated on the RSTPreID dataset, achieving a Rank 1 accuracy of 33.9%, Rank 5 accuracy of$\mathbf{5 6. 2 0 \%}$, and Rank 10 accuracy of$\mathbf{6 6. 5 \%}$.
Fields of Study & Subjects
Video Surveillance and Tracking Methods
Advanced Image and Video Retrieval Techniques
Human Pose and Action Recognition
Publication details
- Journal / source
- International Conference on Intelligent Systems for Communication, IoT and Security (ICISCoIS)
- Pages
- 1–6
- Publication date
- 8 January 2026
- Language
- EN