Error when training tiny YOLO on early stopping

[Jaen0302]

Error when training tiny YOLO on early stopping

[AntonMu]

Hi Jaen,

I haven't done too much testing for tiny Yolo.

Do you have a fork you can share or a colab? Along with some minimal example to reproduce the error?

[Jaen0302]

Hi Anton,

Thank you for such a quick reply! I do have a colab, with the fork in it. Here's the link: https://colab.research.google.com/drive/18r0372CW-Ojo9aON9OyvFQh1joYFDcwu?usp=sharing

As for some minimal example to produce the error, I'm sorry that I do not have it since it only appears after 80 or so epochs. The number is different every time I try it, and do not have a clue about the solution. I do have a picture for the error though, as shown below. For errors #3&#4, it comes out whenever I load tensorflow.

Thanks for your help!

Sincerely,
Jaen

[AntonMu]

Re the training "error" I think you can ignore that. Early stopping just means that the loss has not improved in x number of epochs and thus it is done training.

If you can run inference on it, it means that it is trained successfully. How are the predictions for those pictures that work? Do they seem ok?

I would try to debug the images you are having trouble with. Are they always the same images or do they change?

[Jaen0302]

Hi Anton,

Got it! Thanks a lot~ I'll ignore that error.
Yes, they are always the same images that have problems. I guess it has to do with images that have no detection since it error always comes like this no matter how many times I run it:

Really appreciate your help!

Sincerely,
Jaen

[Jaen0302]

Hi Anton,

I fixed the error somehow, guessed it was just probably because I moved something in the yolo.py. I re-downloaded and ran it again and it somehow worked...

Thank you for your time and help!

Sincerely,
Jaen