Transformer and Residual Connection based Signet for Offline Signatures Verification

We propose two enhancements to the Signet model for offline signature verification. While Signet performs well on Bengali, Hindi, and UTSig datasets, our modifications address its limitations on others. TranReSignet-16-128 outperforms Signet on the CEDAR dataset, and ReSignet improves AUC by 6% on GPDS-150. These adjustments demonstrate that tailored models can enhance accuracy and efficiency, offering effective alternatives for datasets where Signet under performs.