Warnings while running the test scripts about rams HOT 5 OPEN

Hi @pankhurivanjani!

mmmm here something is really going bad :)

Could you give more information on what you've done and what you're trying to achieve? It really seems that the Checkpoint class is not able to resolve some objects.

from rams.

Hi @EscVM,
We're not quite sure either. We've been running multiple experiments, trying to figure out what could be going wrong.
And to further complicate, we don't have direct GPU access, we are having to submit our jobs through HTCondor, a High Performance pipeline for submitting our job to GPU and then observe the errors only via logs.

About 2 days back, we had tried training your model without changing the code for 50 epochs. But for some reason, there was no new checkpoint created. We weren't quite sure about this.

Late yesterday, we tried running only 3 epochs, but we changed the value passed to evaluate_every inside trainer_rams.fit() function from 400 to 1. This resulted in getting a checkpoint file. Upon executing test using this modified checkpoint, we obtained a prediction and it didn't give any of these warnings.

Would you know anything around why this could be happening? We are not able to understand if it's due to the evaluate_every parameter or something else.

from rams.

Forgot to add the image, this is our result obtained after 3 epochs and evaluate_every =1

from rams.

As you can check in line 165 of training.py script, evaluate_every simply triggers the test procedure. Moreover, if the test PSNR is lower than the previous test, the code saves the network's weights. That's it.

Why don't you iterate with the code with a Colab session? You could quickly check if you're missing something.

from rams.

I don't mean to hijack this issue, but I noted another issue where the user is not using a GPU even. I can confirm Google Colab can be used for training, with 25 epochs taking approx 4-5 hours

from rams.