Closed mucunwuxian closed 8 months ago
Hello! Thank you very much for your words.
Yes, we indeed conducted experiments with the FDS technique as well, but none of our final models incorporate it. We did not observe significant improvement with its use, and the computational cost during the training stage is quite substantial, especially with large datasets. Consequently, we decided against it. I think we should remove it from the code. Thank you.
I understood clearly! I was given very valiuable knowledge. Thank you for your thorough reply!
P.S. I posted the above content on my tech blog. https://medium.com/axinc/multi-input-transformer-for-age-and-gender-estimation-%E5%B9%B4%E9%BD%A2%E3%81%A8%E6%80%A7%E5%88%A5%E3%82%92%E6%8E%A8%E5%AE%9A%E3%81%99%E3%82%8B%E3%81%9F%E3%82%81%E3%81%AE%E3%83%9E%E3%83%AB%E3%83%81%E3%82%A4%E3%83%B3%E3%83%97%E3%83%83%E3%83%88%E3%83%88%E3%83%A9%E3%83%B3%E3%82%B9%E3%83%95%E3%82%A9%E3%83%BC%E3%83%9E%E3%83%BC%E3%83%A2%E3%83%87%E3%83%AB-8b77aa8c6dbc
Thank you for your great article and repository. I was impressed.
Please tell me about the code below. https://github.com/WildChlamydia/MiVOLO/blob/main/mivolo/model/mivolo_model.py#L283
LDS is mentioned in the paper, but FDS seems not to be mentioned. However, the code in the link above shows that FDS is applied. Did you apply it on a trial basis while studying? I looked for
_fds_forward
but couldn't find it.