Open Inferencer opened 4 weeks ago
- 可视化工具 TensorBoard不错,用法也很简单
- 针对特定人物进行训练,我觉得在通模的基础上 fine-tune效率更高一些
I would fine tune on the general model if I could start at clip training or frame training 128 by just adding the extra inputs such as csv and video along with it's current dataset is that possible then?
When I tried person specific training last year with just two short clips it was very fast and I would like to keep this down to a couple of hours rather than a day worth of training and that would be optimal for me but if there is any negative effects that come come from using person specific than maybe I need to change the code.
Yes tensorboard is what I need but last year I failed to correctly implement it, I was hoping somebody had some working code already used in their DINet that I can have
As the title says I need a visualization of the convergence while training, I haven't trained a DINet model in a year and there seems to be a lot more people attempting to train.
Also if anybody has attempted person specific training I would like some tips on source footage duration