Search papers, labs, and topics across Lattice.
University of South Alabama Corresponding author. Email: crumeike@crimson.ua.edu, University of South Alabama
2
0
4
0
Transformer-based detectors can significantly improve damage severity estimation by incorporating ordinal-aware supervision, even surpassing CNN-based models in ordinal consistency.
Forget fancy architectures: simply switching from Adam to SGD can boost Vision Transformer performance by +25 to +38 F1 points in post-tornado damage recognition.
