kongbia
kongbia
Is now the problem solved
I think it may be an encoding problem by ```json.dumps``` and have solved it by adding ``` class DateEnconding(json.JSONEncoder): def default(self, o): if isinstance(o, datetime.date): return o.strftime('%Y/%m/%d') ``` in ```vot/document/__init__.py```...
I used Toolkit version: 0.4.2, python 3.6, pytorch 1.3, cuda10+cudnn7 and the tracker is the official oceanplus
I see. Thank you very much!
感谢！
> @kongbia > This codes must be remained because pos means positive samples and pos exists during training also. but in siammask_sharp.py line 65, there is no pos input during...
> > 谢谢，方便的话能否分享下您的ddp代码？我写的ddp不知道哪里出了问题，训练的过程慢的离谱2021年8月4日 18:09，Fish Jack _**@**_.***>写道： 你好，请问你用ddp4卡训练花费多长时间？ 我尝试用ddp训练，速度提高并不明显 我ddp4卡2080ti 32batch训练5天左右，训了800/400个epoch，其他的参数和论文相同 —You are receiving this because you commented.Reply to this email directly, view it on GitHub, or unsubscribe.Triage notifications on the...
> 你好，请问你用ddp4卡训练花费多长时间？ 我尝试用ddp训练，速度提高并不明显 我用ddp 4卡 num_work=4训练，和不用ddp 4卡 num_work=16花费的时间差不多