metadata
license: apache-2.0
3️⃣ KAT-ReID – Market-1501
Model Details
- Model name: KAT-ReID (Market-1501)
- Task: Person Re-Identification
- Dataset: Market-1501
Model Description
This model evaluates whether KAT’s ImageNet gains translate to standard person ReID benchmarks with minimal occlusion.
It serves as a generalization stress test for GR-KAN-based transformers.
Training Data
- Identities: 1,501
- Images: 32,668
- Cameras: 6
Evaluation Results
| Metric | Score |
|---|---|
| mAP | 78.8 |
| Rank-1 | 91.1 |
| Rank-5 | 96.9 |
| Rank-10 | 98.2 |
While competitive, the model does not surpass ViT-based SOTA, highlighting the task-specific nature of KAT benefits.
Limitations
- Lower global discriminative power vs ViT baselines
- Performance sensitive to patch overlap and local branch weighting
Citation
@inproceedings{umair2025katreid,
title={KAT-ReID: Assessing the Viability of Kolmogorov--Arnold Transformers in Object Re-Identification},
author={Umair, Muhammad and Zhou, Jun and Musaddiq, Muhammad Hammad and Muhammad, Ahmad},
year={2025}
}