Ignite: [Feature Request] Neural Style example

[vfdev-5]

Following #202 idea is to port fast neural style example from official pytorch examples.

[anmolsjoshi]

@vfdev-5 , I'm interested in helping out with this issue.

I've been wanting to contribute to open source for a while, and this would be a good start.

[vfdev-5]

@anmolsjoshi sounds good! Feel free to post any questions here, e.g if you have problems with implementation or encounter some limits of the API.
You can inspire from other examples on how to port examples from official pytorch examples to ignite.
We have smoke tests for the examples, so once your code is locally running and gives similar results as the original official one, then you will need to integrate it into .travis.yaml.
When you are done with the code in your fork in a branch (e.g. named neural_style_example), then please submit a PR, we will review it and if everything is OK, CI passes, then it will merged into the master and it will be your first contribution to Ignite :)

[anmolsjoshi]

@vfdev-5 Thanks for the input! I've already gotten started on this project.

[alykhantejani]

@anmolsjoshi thanks for taking this up, and welcome to open source!

[anmolsjoshi]

@alykhantejani thanks!

@vfdev-5 I have been able to reproduce results of the original example, it is working locally. I am unfamiliar integrating the code with .travis.yaml. Any guidance on that?

[vfdev-5]

@anmolsjoshi good news! So, about .travis.yaml. We are just providing a sort of smoke tests for examples = run example in a tiny configuration and check that returned code is 0. For example,
https://github.com/pytorch/ignite/blob/c4563ddc67bee03fc616e87b84abd9b5e17e21cc/.travis.yml#L36
In your case you can add after the line with RL examples a command to run your example: (smth like that)

- python examples/fast_neural_style/neural_style.py --epochs 1

When you send a PR, it will be executed with travis and it will give you the result of the run.
While in the PR you can make any number of modifications of the code you want, so if it does not work from the first time you can fix it with another commit. When PR is accepted we will squash all in a single commit, no problems.

[anmolsjoshi]

@vfdev-5 thanks for the help! this code requires MSCOCO dataset and a few images downloaded before, to run correctly. Am I supposed to incorporate that into the .travis.yaml file? Sorry if this is a dumb question.

[vfdev-5]

@anmolsjoshi

this code requires MSCOCO dataset and a few images downloaded before, to run correctly.

If it possible to run training on Cifar10 instead of MSCOCO would be better. I mean if --dataset option could recieve a path and a str cifar10 and then configure the dataloader with CIfar10 dataset. What do you think ?

Could you please also send a PR, we will consider it as Work-in-progress to take a look on your code and iterate on what can be done.

Am I supposed to incorporate that into the .travis.yaml file?

Idea is to check if code is just running. Result correctness is not checked. As you have two parts: training and eval, the training can be done on a smaller dataset or even a fake dataset and evaluation can be also done a fake image. So maybe we need to tweak a little --content-image with acceptable "random" value.

[anmolsjoshi]

@vfdev-5 Thanks for the help! I've submitted a pull request. Let me know if it needs any corrections.

The Travis CI build has been going on for over 40 minutes. If you can, could you stop that build? I'm not sure what the issue is. It works correctly on my local machine.

[anmolsjoshi]

@vfdev-5 Can this be closed?

[vfdev-5]

Yes, @anmolsjoshi Thanks!

Solved by #222