training yolov3 with single class

[Shanthika]

Hello,

I was trying to train yolov3 with a single class and got this error:

parse_yolo: Assertion `l.outputs == params.inputs’ failed.

I searched and found out that it is due to wrong filter value. I have given the filter value as 18(because its for a single class) and it's not showing error at line 603 but at 699 and 776.

Please tell me what value I'm supposed to enter at those lines.

[AlexeyAB]

@Shanthika Hi,

Show screenshot of output with this error.

[Shanthika]

Here's the link to the screenshot

https://drive.google.com/open?id=16uThRxweTm4QY0S3xNk0UxS0A0VaBjoa

[AlexeyAB]

You google-drive requires permission. Better drag-n-drop your image here to your message, or press selecting them.

[Shanthika]

Here's the link with permission: https://drive.google.com/open?id=16uThRxweTm4QY0S3xNk0UxS0A0VaBjoa

Also attachment:

[AlexeyAB]

@Shanthika

Rename your cfg-file to the yolov3-obj.txt and drag-n-drop to your message.

Check that parameter classes=1 for each of 3 [yolo] layers in your cfg-file.

[Shanthika]

Okay, I checked the classes value. I had skipped changing the value of class at line 696. I corrected it and it is working now. Thank you for your help.

I had a question though. We are working on detection of objects in 360degree videos and intend to use yolov3 to detect objects in equirectangular images. So, is it possible to train few of the classes with new images and keeping the other classes intact?

[AlexeyAB]

So, is it possible to train few of the classes with new images and keeping the other classes intact?

You should train from the begining, if you changed the number of classes.

[Shanthika]

Okay, thank you for your time.

[kalikhademi]

I had the same error and I checked that all classes =1 . I have dropped my yolo_obj as txt file. I still have this error. Can someone help me plz!

yolo_obj.txt