Is it possible to force the algorithm be deterministic ?

[0wnrepo]

Each time I retry the algorithm, it produces a different output. Is it possible to set an initial seed to keep it deterministic aka same output each run ?

Regards

[jcjohnson]

You should get deterministic output if you call the manualSeed function at the start:

https://github.com/torch/torch7/blob/master/doc/random.md#manualseednumber

[0wnrepo]

Thank you ! That's what I've been looking for.

[nthverse]

If I am doing a frame to frame video will this help have the results seem smoother from one frame to the next? or is it the case that it'd have no effect at all.

[nthverse]

I can run a test if you are curious, I just need a bit of info on how and when I use torch.manualSeed(123) .

[0wnrepo]

It will have the effect desired. For example I tried to generate an artistic image based on a portrait of my girlfriend, the first time the eyes were completely white, another try one eye was black, while the other had some black spots - using the same photo. Using the manualSeed you don't have to worry about this.
Because the frames will be different though, the output will be a bit different but not so much as if you were using de default settings - that is random seed.

[nthverse]

do I run this once per torch session

edit it does seem to be the case

[jcjohnson]

You can just use the -seed flag: https://github.com/jcjohnson/neural-style/blob/master/neural_style.lua#L43

I completely forgot that this flag existed!

[nthverse]

oh wonderful. thanks jcj