Dear Dr. Lee, Dr. Gong, and Dr. Cho,
I’m HSZ, reproducing the experiments in your CVPR 2025 paper. Thank you for open-sourcing the resources.
While using your provided data and checkpoints, my results on TVSum match your paper well. However, there’s a noticeable gap when running on SumMe. I’ve verified the directory structure, hyperparameters, and checkpoint alignment as per your docs.
Could you share insights on SumMe-specific steps (e.g., KTS parameters, annotation aggregation) that might explain this? Your guidance would be much appreciated.
Best regards,HSZ
Dear Dr. Lee, Dr. Gong, and Dr. Cho,
I’m HSZ, reproducing the experiments in your CVPR 2025 paper. Thank you for open-sourcing the resources.
While using your provided data and checkpoints, my results on TVSum match your paper well. However, there’s a noticeable gap when running on SumMe. I’ve verified the directory structure, hyperparameters, and checkpoint alignment as per your docs.
Could you share insights on SumMe-specific steps (e.g., KTS parameters, annotation aggregation) that might explain this? Your guidance would be much appreciated.
Best regards,HSZ