How to obtain the test results? about eli HOT 14 CLOSED

You can read the last few lines in the log. Hope you are using the provided scripts as is.

from eli.

Thank you. This is my log. However, I find the results are very low. I directly perform run.sh

.

from eli.

Please share your complete logs so that we can debug. I wonder whether the base models are loaded properly. Thanks.

from eli.

log.txt

This is my log

from eli.

./output/10_p_10_ft/model_final.pth is not being loaded. This is evident from the lines "[04/17 11:57:02] d2.evaluation.pascal_voc_evaluation". Can you please share ls ./output/10_p_10_ft?

Also, please find my log for your reference:
log.txt

from eli.

Thanks. I do not change your code. Could you give me some advice to reproduce your results?

from eli.

Is your run.sh correct?

from eli.

Yes, it is correct. Please respond to my question: Can you please share "ls ./output/10_p_10_ft".

Have you downloaded and extracted the models as explained in the third point here?

from eli.

Yes. I have downloaded the models. I observe that your code needs to use these models.

from eli.

1. What is your version of fvcore?

2. Also, do you see any errors while the code loads the models? These errors may appear only on the screen, it might not necessarily be logged to file.

3. Can you show the contents inside 10_p_10_ft?

from eli.

The version of fvcore is 0.1.1.dev200512

from eli.

The 10_p_10_ft is the same as my download. It does not change.

from eli.

Thanks for sharing these. As I mentioned before, the issues is in loading the model parameters. That is why the accuracies are very low (you can do a side-by-side comparison to my logs).

I wonder whether it is a file permission issue. Instead of creating shortcuts, can you copy paste the contents from the extracted models into "output" folder? Also, is there any extra message that comes up in the standard system-output when the following log comes: "fvcore.common.checkpoint INFO: Loading checkpoint from"

(I am bit busy during the day, will get back later.)

from eli.

Hello, please reopen this incase your issue is not yet resolved. #6 might have been the reason.

from eli.