FrankenGAN with chordatlas - bikeGAN root not found

[twak]

FrankenGAN with chordatlas - bikeGAN root not found

[twak]

Hi, thanks for reporting this. Yes, this sounds like the correct workflow.

The png files with the strange names are the inputs that chordatlas writes out for bikegan to process. Once the go file is written, bikegan should automatically i) process these to outputs, ii) delete the go file and inputs, and iii) write the output to the bikegan output folder.

Some questions (if you don't want to post then answers on a public forum, email me:

1/ What is the (exact) bikegan root that you set in chordatlas?
2/ Which operating system are you using?
3/ After the inputs are written, is there any additional output from bikegan?

[twak]

One more thought I had while cycling to work: you must have to docker (or nvidia-docker) running as the same user as chordatlas. (i.e. you shouldn't be running docker with sudo). You can add your user to the same group with these instructions. You then have to restart your computer.

This is because bikegan uses python's watchdog to listen for new input, and it doesn't work across users under docker. This probably means the docker version won't work under windows :(

[twak]

No comment back for a month; closing issue :(

[leekkww]

Sorry about the delay - was on extended vacation. I have a suspicion that the error is due to the machine that I am constrained under, and I am trying to move the code over to another machine. Will let you know if a problem still exists!

[leekkww]

I have moved to a system with no constraints but this problem seems to persist - would you be fine if we reopened the issue?

I am not running docker with sudo - I am running as kwlee (me), which I have put into the docker group.

To answer your questions:

1. The exact BikeGAN root that I am using is (and I have read / write permissions in this directory) /home/kwlee/bikegan_temp. (I have also tried /home/kwlee/bikegan_temp/ with the extra slash in the end, if that makes a difference)
2. My system is Ubuntu 14.04.5 LTS.
3. there are no additional outputs from bikeGAN beyond the following:

model [BiCycleGANModel] was created
all nets up
[network super10 is awaiting input]

If I run ps aux | grep "test_interactive.py", I see the following:
1000 25356 4.8 3.5 49243592 2339560 pts/0 Sl+ 15:33 0:09 python /home/user/bikegan/test_interactive.py
which means that user 1000 is running the script, not me. Does that make a difference?

I also have a feeling that this may be related to this issue that I opened on BikeGAN. Perhaps my solution of using relative path names is causing more problems than I expected.

[twak]

Hey, hope you had a great holiday! Yes it would be great to get to understand what is happening.

I have a clean ubuntu install at home; I'll try to reproduce your workflow tonight.

[leekkww]

I ran the same system without nvidia-docker according to instructions here and it worked perfectly! Perhaps it is a problem with permission settings with docker then.

[twak]

Great that you got it to work! If you use the project for anything, I'd love to know - it all helps with future funding etc..

I found a couple of issues with the docker container, so I rebuilt it...I'm not certain these were causing your problems though.

Two thoughts:

• I notice that you omitted the pwd bit when starting the docker container - when I tried this, docker didn't deal well with relative paths
• The input and output folders need to exist before you start the container. When the container starts it should populate input with the different input points for the networks. I updated the instructions to make this clear.

It would be great if you can test my new image. Create a new temporary director, cd into it, and run

mkdir input output
nvidia-docker run -v $(pwd)/input:/home/user/bikegan/input -v $(pwd)/output:/home/user/bikegan/output -it --rm twak/bikegan

I can run docker with sudo fine now, so I don't think this is the problem.

[leekkww]

I tried the new image and I'm still running into the same issue described here, where I get many permission denied errors. If I remove the $(pwd) bit, that doesn't give the permission denied errors, but then the script hangs as describes. Let me know if any more information about my system would be helpful!

[twak]

thanks for trying - do both the input and output folders exist before you start the image? If I don't create input and output I get similar error messages:

PermissionError: [Errno 13] Permission denied: './input/door textures'

Then it creates the folders with the wrong permissions (root). ls -la gives:

drwxr-xr-x 56 twak twak 4096 Feb 21 18:34 ..
drwxr-xr-x  2 root root 4096 Feb 21 18:45 input
drwxr-xr-x  2 root root 4096 Feb 21 18:45 output

Further runs cannot write to these folders (and I guess neither can chordatlas).

if I do sudo rm -rf input output, then mkdir input output (from your non-root account), it works.

In other news, if I run without pwd (nvidia-docker run -v input:/home/user/bikegan/input -v output:/home/user/bikegan/output -it --rm twak/bikegan) it doesn't create input, or output, in the current directory, so I don't know where the output would go!

[twak]

I'm going to assume that the previous reply fixed the problem. Shout if this is still an issue...I can't reproduce it any other way.