pytorcher

Bugs in the code?

1

hi, @WongKinYiu @AlexeyAB I have questions for the code : 1 . in the training part https://github.com/WongKinYiu/PyTorch_YOLOv4/blob/3c42cbd1b0fa28ad19436d01e0e240404463ff80/train.py#L242 you resize the image by a random scale, but the correnponding target annotations...

hi，@WongKinYiu in https://github.com/WongKinYiu/PyTorch_YOLOv4/blob/3c42cbd1b0fa28ad19436d01e0e240404463ff80/test.py#L258 is this line redundant? a little bit strange？

Question about generator training

2

hi: Have a question , in https://github.com/AliaksandrSiarohin/monkey-net/blob/0c8aa4ea458a79717d9016a36a3d86d10a166aea/train.py#L48 for the generator training, when computing the generator loss, it seems the discriminator parameters are not frozen, while for discriminator training, https://github.com/AliaksandrSiarohin/monkey-net/blob/0c8aa4ea458a79717d9016a36a3d86d10a166aea/train.py#L70, the...

hi, thanks for your great and amazing work, in the code https://github.com/AliaksandrSiarohin/first-order-model/blob/ca49071ce60051200f58f03a0e4e65e675a54e27/modules/model.py#L101 I am confused, why compute the distance between the input coordinates and the control points? ` result =...

training code

2

hi， when will you plan to release the training code?

Implementation detail about detectoRS

1

hi, in the implementation of detectors_resnet.py, https://github.com/open-mmlab/mmdetection/blob/3a9f5eff39735871c55643927f5506ab5af2127a/mmdet/models/backbones/detectors_resnet.py#L300,the rfp feature seems not to be added to the first stage of the resnet, can explain the reason? and the paper seems not...

你好，请问有计划将你的方法集成到mmdet里面去吗

3

您好，请问： 1.有计划将论文方法加到mmdet里面去吗？感觉加入进去后，更方便和其他方法组合，另外也可以比较公平的来对比各个方法的性能 2.有个问题，如果一个proposal和两个gt boxes匹配上了，并且这两个gt boxes的标签都是同一类，比如说person，请问训练时，这个proposal 的label就是两个1，是吗?(这里的1代表person类的标签) 3. 另外，文章中的refinement module，是不是和cascade的思想是一样的呢？拿第一次得到的预测结果，然后再做roi pooling后，再做一次分类和回归？我看代码实现由one stage和two stage两种，是不是refinement 模块只能用在two stage上呢？