yml problem

[QZ-cmd]

Hello, I would like to replace FEATURE_LAYER yml, how to determine the FEATURE_LAYER profile, such as yolov3 and mobilenetv3

[foreverYoungGitHub]

Seems I forget update the relative config files and pretrained model file. But an example yml can be find at here: https://github.com/ShuangXieIrene/ssds.pytorch/blob/master/experiments/cfgs/tests/test.yml.

You can check the config document for this lib at here: https://foreveryounggithub.github.io/ssds.doc/config.html.

Let me know if you still have any problem

[QZ-cmd]

Thank you for your reply, the new version of yml is different from the old version of yml, I also looked at config.html, but please forgive me still do not know how to set such as SIZES or ASPECT_RATIOS, and so on, if you can add a few more yml examples, I will thank you very much, I need yolov3 / v2 +mobilenetv2 yml.

[wiebevr]

Hi all,

First of all thank you for providing an implementation of that many detectors / architectures!

I'm having the same issue figuring out what values should be filled in the FEATURE_LAYER configuration parameter. I'm trying to train mobilenetv1/2 + yolov3.
What I've come up with now is
FEATURE_LAYER: [[0, 1, 2], [256, 512, 1024]]
with the reasoning that it needs the number of ouput channels for each feature layer that is used in the YOLO head, and from the code it seems there are only 3 feature layers in the mobilnetv1 implementation, hence the 0, 1, 2.
This doesn't work however, has anyone gotten this new version working?
An example configuration file could indeed be helpful!

Thank you for the help!