Yolov7 Architecture in Research Paper

WongKinYiu / yolov7

Implementation of paper - YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors

GNU General Public License v3.0

13.37k stars 4.22k forks source link

Yolov7 Architecture in Research Paper #762

Open thilinawee opened 2 years ago

thilinawee commented 2 years ago

Hi, When I went through the research paper of yolov7, I came across this diagram which describes the architectural differences between some other networks and yolov7 itself.

I wonder if anyone can explain to me the meanings of these notations. For example, what does 3x3, 2c, 2c, 2 stand for? I understand that 3x3 means a kernel_size = 3,3. However, I am not sure about the rest. Thanks ~Thilina

WongKinYiu commented 2 years ago

kernel size, input channel, output channel(, group).

thilinawee commented 2 years ago

Thanks, @WongKinYiu for the quick reply. Does group mean the number of groups in the convolution(group convolutions) or the number of such sequential blocks?

WongKinYiu commented 2 years ago

They are equivalent, you could take a look ResNeXt paper.

thilinawee commented 2 years ago

I sketched the yolov7 architecture according to what I understood. Could you please point out where is E-ELAN block with group convolutions? Are the group convolutions used only for training? Thanks in advance.