-
Notifications
You must be signed in to change notification settings - Fork 43
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Error in dataset path verification #7
Comments
Hi @jmsteitz , thanks for the heads up. Let me look into this. On which day did you clone the repo? I merged a few PRs since initial release. |
I cloned it two days ago, so I should have the latest version. |
Ok that's good -- let me run the script here on my local machine and see if I can reproduce it, can you give me 1-2 days to get back to you? |
Did you save the output log (stdout) as the download scripts executed? were there any previous errors? |
Could you also paste here the exact commands you ran here from the download_scripts readme? Did you upload the Cityscapes, BDD, and WildDash files to your machine after a Chrome download? |
I wasn't able to reproduce this error. My guess is that the directory structure may not be correct on your machine. Could you show me a graphical representation of the directories you've configured (as done here), and let me know what you've named I did notice that |
MSEG_DST_DIR is set to '/fastdata/jmsteitz' on my machine. The folder structure looks like this right now:
Cityscapes was already available on our file server, but I indeed downloaded WildDash and BDD with Chrome on macOS and then uploaded them to my machine. But they still seemed to be handled fine by their scripts. I'm afraid I don't have any output logs but I don't remember any errors either. |
Hi @jmsteitz, those directories all look correct, which is good. I may add output logs directly to the instructions for future users. Since ade20k-151 seems to be throwing issues for you, can we check some paths in there? for example, do these files exist in your directory?
|
Could you let me know the exact filepath it's breaking on for you? This part of the script ran fine for me. |
Yes, both paths exist. But I guess, it's not specifically ade20k-151 but a more general problem, because when I skip the ADE20K dataset in the verification script it still throws the error (but with BDD then). |
Got it, thanks for your patience while we try to figure this out. Thats good those exist. This part of the script is just looking for paths, so we need to find out which line it breaks on for you, so we can see if a path is really missing. Do you mind setting a conditional breakpoint and stepping through it, or adding a print statement in the loop? Also, did you set the variable in mseg/utils/dataset_confif.py? Hoping we can avoid the situation where you would have to rerun everything and check output logs for an error along the way. |
It's directly breaking on the first set, which is
and where we already made sure it exists. It seems, that we somehow end up with an all zero |
Got it, ok that's helpful. What does
return to you?
|
Is your
either Can you also run |
The unit tests all pass, my cv2 version is 4.2.0 and I got the exact same output when printing the labels as you do. |
Got it, that's good. Do you mind setting a breakpoint before it crashes using |
Well, So I had a look into
The first list of vertices to build
so it generates an empty mask and leads to the numpy error in the end. My Pillow version is 7.1.1 by the way. Does your first |
If I append two print statments after that line in mask_utils_detectron2.py:
I get the following output. Could you check the diff? Thanks again for your patience here. |
As mentioned in my earlier post, When I insert Could you uncomment |
Thanks for adding the minimal code example -- a line segment shows up in my array, unlike on your machine:
|
Yes, it seems to be a Pillow bug, see here. My colors in the overlay produced by the Visualizer are different than yours, though. Is that a problem? Now, the verification script runs until |
Hi @jmsteitz, thanks for all this feedback. I've pushed several changes that incorporate your polygon vs. line fix for horizontal lines. |
Hi @jmsteitz , let me know if you run into any further issues. I'm closing this for now since I merged fixes in two PRs, but feel free to re-open it if your issue is not resolved. |
When I try to run the verification script
I get the following error:
Additonally adding 'ade20k-151', 'ade20k-150', and 'ade20k-150-relabeled' in line 58 to skip them, leads to the same error thrown for the BDD dataset. Thus, the error doesn't seem to be dataset specific.
I'm running python 3.7.6 and numpy 1.18.2.
The text was updated successfully, but these errors were encountered: