sebastianvitterso / master-sau

Sau og master
GNU General Public License v3.0
6 stars 2 forks source link

What is the structure of Yolo V5 Generic Fusion? #3

Closed 283pm closed 2 years ago

283pm commented 2 years ago

First of all, thank you for this great program.

I am a beginner and have some questions.

Regarding the network structure of generic fusion, is it correct that the RGB and Ir processes are fused just before the head?

Also, I was looking at the Tensorboard graph and I am a little confused. Is the output from SPPF being input to the first layer simply because the same processing that is used for SPPF is used for the output from SPPF as for the first input?

sebastianvitterso commented 2 years ago

Regarding the network structure of generic fusion, is it correct that the RGB and Ir processes are fused just before the head?

This is correct, yes. To be more precise, the fusion happens right after the backbones, and all the skip paths also contain a fusion layer to allow for them to work. See our thesis figure below:

image.


Also, I was looking at the Tensorboard graph and I am a little confused. Is the output from SPPF being input to the first layer simply because the same processing that is used for SPPF is used for the output from SPPF as for the first input?

I'm not quite following, but that might be just because it's so early in the morning. Any chance the figure answers the question?

283pm commented 2 years ago

Thank you for your fast reply.

My problem is solved by your great image.

Thank you.

sebastianvitterso commented 2 years ago

Good to hear! Good luck with your work.