Inference in editor is not the same as runtime

[Sohojoe]

When running the trained model in the Unity Editor - you may see different results between runs and from Build->Run

Workaround

• Build->Run - i.e. create an executable project to view results

Reproduced in 2018.4, 2019.1 2019.2beta, 2019.3 alpha

[joanllobera]

Hi Joe, I have trained the styletransfer002 scene, with the backflip, on 2019.1, twice. In editor it does not work. However, in a build it does neither. Do you have any idea why this might be happening?
I attach the model resulting from the second training, compressed, in case you want to try it with your build.

. myBackFlip_myBackFkipTransfer002-0.zip

by the way, I am doing all this on windows 10.

I will try to train and then see the results in 2018.4, maybe I am a bit more lucky, and I can start looking further into the details. Any advice welcome! Regards, Joan

[Sohojoe]

do you have the latest code? https://github.com/Sohojoe/ActiveRagdollStyleTransfer/commit/6a3399c0db4395f4f8c954fbcecffab1dbf7560e

Sorry for not getting back earlier - i've been traveling. I do now have marathon-envs (my dev) updated to the ml-agents v0.8.2 - next i will implement this into style transfer.

I will check you trained model and get back!

[joanllobera]

Thanks for the update! I will check.this commit first thing whem I.am back in,front.of.a computer

Joan

On Fri, 26 Jul 2019, 20:27 Joe Booth, notifications@github.com wrote:

do you have the latest code? 6a3399c https://github.com/Sohojoe/ActiveRagdollStyleTransfer/commit/6a3399c0db4395f4f8c954fbcecffab1dbf7560e

Sorry for not getting back earlier - i've been traveling. I do now have marathon-envs (my dev) updated to the ml-agents v0.8.2 - next i will implement this into style transfer.

I will check you trained model and get back!

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/Sohojoe/ActiveRagdollStyleTransfer/issues/5?email_source=notifications&email_token=AA5H4TAIYE7I3RVY5AEYAPDQBM62HA5CNFSM4IDBYTK2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD25LZ5I#issuecomment-515554549, or mute the thread https://github.com/notifications/unsubscribe-auth/AA5H4TEFWG2HC7OVBLDBYDDQBM62HANCNFSM4IDBYTKQ .

[joanllobera]

hi Joe, I am back on track into looking into this in more detail.

Is the update of the marathon environments to ml-agents v0.8.2 online? The last commit in your repo seems to be from 8 months ago. I see that you committed something to the activeRagdoll repo two weeks ago. I will try that out, see if now I can get this to work on my computer, and let you know how I progress. Cheers, Joan

Missatge de joan joan.llobera@gmail.com del dia dv., 26 de jul. 2019 a les 22:16:

Thanks for the update! I will check.this commit first thing whem I.am back in,front.of.a computer

Joan

On Fri, 26 Jul 2019, 20:27 Joe Booth, notifications@github.com wrote:

do you have the latest code? 6a3399c https://github.com/Sohojoe/ActiveRagdollStyleTransfer/commit/6a3399c0db4395f4f8c954fbcecffab1dbf7560e

Sorry for not getting back earlier - i've been traveling. I do now have marathon-envs (my dev) updated to the ml-agents v0.8.2 - next i will implement this into style transfer.

I will check you trained model and get back!

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/Sohojoe/ActiveRagdollStyleTransfer/issues/5?email_source=notifications&email_token=AA5H4TAIYE7I3RVY5AEYAPDQBM62HA5CNFSM4IDBYTK2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD25LZ5I#issuecomment-515554549, or mute the thread https://github.com/notifications/unsubscribe-auth/AA5H4TEFWG2HC7OVBLDBYDDQBM62HANCNFSM4IDBYTKQ .

--

Joan Llobera - http://joan.llobera.name joan.llobera@gmail.com

[Sohojoe]

Hi Joan, yes - I have marathon envs updated to ml-agents v0.9 - here

I started updating the style-transfer repro but went down a bad path - so back working on that now.