Kunyang Zhou
Kunyang Zhou
Thanks! Your data processing code about Human3.6M provide three kinds of coordinate which is shown in the picture, If I want to predict 3D coordinate from 2D coordinate which kind...
> Hi, > > I am using CUDA10.1. > > I do not know how to set a correct environment. If I run "make install", it will automatically have 'torch...
> > Pytorch >=1.8 needs CUDA>=10.2 > > I think you should choose correct pytorch version which can be compiled with CUDA==10.1 > > By the way, the setting in...
> 可以的。做过了没问题。 谢谢，有个问题想问您一下，就是我使用one-stage 检测器时候发现无标签部分教师模型无法生成伪标签，然后找到原因是因为置信度太低了，而降低前景的阈值又会导致产生噪声伪标签，想问一下这个问题您遇到过吗，谢谢
> 我做消融的时候也没有遇到。。。。你的在有标签的数据上面正常吗，可以用训练出来的权重inference一下试试。。。如果即时训练轮数很大了也不正常的话那应该就是loss function等设计的问题了，如果正常的话那就挺奇怪的了。。。刚开始训练的时候没有伪标签是挺正常的 您是用的什么模型？我是使用的FCOS，但是整个训练过程都没用伪标签产生，不知道哪边出了问题
Hi,How can I find the savd model? I find a folder named output,but in the folder i can not find .pth. 
> @qdd1234, I used register_coco_instances to register my custom dataset but when I pass it to train_net.py it shows that my dataset has not been registered. Could you show me...
> @qdd1234 But you are using train_net.py to train, right? > Thanks for the quick reply Yes
I directly changed the configuration of the VOC data set in Detectron2, including the VOC data category, the original VOC path, and the file name when reading the TXT file...
Hi ，If I want to train the model on my own dataste，How can i do？