Effect of Dimensionality Reduction on Cross-Domain Performance in Person Re-Identification Kişi Yeniden Tanimada Boyut ?Indirgemenin Çapraz Alan Başarimi Üzerindeki Etkisi
34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Türkiye, 7 - 10 Temmuz 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/siu71813.2026.11636396
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: cross domain generalization, dimensionality reduction, person re-identification, principal component analysis, vision transformer
- Yıldız Teknik Üniversitesi Adresli: Evet
Özet
In image-based person reidentification systems, the identification process relies on the similarity computation between feature vectors extracted from query images and existing vectors in large galleries. As the size of the feature vector increases, the amount of data that needs to be kept in memory and the computational cost increase linearly. In this study the compressibility of high-dimensional vectors obtained from vision transformer-based feature extraction models trained on the person reidentification task using principal component analysis and the effects of this process on model performance are investigated. Experimental results show that compressing the feature vectors at rates exceeding 80% causes only a minimal loss in training domain performance. A more remarkable finding is that in cross-domain tests, despite the domain shift problem, the performance drop exhibits a similar behavior to the training domain.