Ethan

What's the difference between Inception Module of TinaFace and Context Module of RetinaFace

1

I noticed that your paper mentioned this, but I didn't notice any difference when I read the code. Looking forward to your reply.

I know you didn't use P2 layer for prediction, in the original YOLOV5 Focus had a lower sampling rate of 2. But the sampling rate of the new Stem structure...

Does your data augment use random scale input?Did you make random clipping directly on the original Image? Looking forward to your reply.

About GPU memory usage

3

If non-local is applied to the low-level feature map, CUDA out of memory will happen.Is this due to the amount of memory required to compute the Attention matrix? Looking forward...

Awesome work. For datasets with larger scale variations than COCO.Are there any good ways to further optimize YOLOF？ For example, is it feasible to deepen Dilated Encoder? Looking forward to...

论文中说以半监督的形式标注了不小于 35×35 像素的人脸，但是我发现公开的数据集中小于 35×35 像素的人脸也有姿态信息，请问这些姿态信息可靠吗？

日志相关

1

1. 异步多线程写日志不会出现日志乱序的问题吗？ 2. 在向文件中写日志需要互斥加锁，那么这种并发有意义吗？和单线程写日志有什么区别？ 以上是面试中问到的，感觉自己回答的不好，有大佬给个解答吗？

torch2trt/main.py 69~71行，此处cv2.rectangle接收的就是左上角和右下角坐标，这里不需要使用xyxy2xywh，并且也不需要 除以gn进行归一化，landmarks同理。 ```python xywh = (xyxy2xywh(det[j, :4].view(1, 4)) / gn).view(-1).tolist() conf = det[j, 4].cpu().numpy() landmarks = (det[j, 5:15].view(1, 10) / gn_lks).view(-1).tolist() ``` 也就是应该和detect_face.py的174~176行一致，但是不知道为什么多写了，修改后为： ```python xywh = det[j, :4].view(1, 4).view(-1).tolist()...

如何对视频文件进行检测？

2

运行video.py文件时，我将capture=cv2.VideoCapture(0)中的0改成了MP4格式的文件路径，但是运行报错了，请问为什么呢？