AirsimE2EDeepLearning code seems to change the tone of my image data

[kutilanak]

Problem description

After completing the tutorial this cookbook gave me, I'm using this cookbook to train my dataset, but the code seems to change the tone of my data image so it can't learn properly.

Problem details

After making my own dataset in this map I made it myself

I tried to train the data using TrainModel.ipynb, but after many iterations, the model it made can't work properly.

So I'm figuring out why, and then I found this:

some of the image tone have changed. The middle line which supposed to be yellow, turned to blue. I think this is the reason why the model it made can't work properly.

the question is, how can I solve this? Thank you :)

Experiment/Environment details

• Tutorial used: AirSimE2EDeepLearning
• Environment used: self made environment. The hawaii one can't support my xbox controller & pxn v3 pro steering wheel.
• Versions of artifacts used (if applicable): python 3.6.9, keras 2.1.2, tensorflow 1.11

[kutilanak]

I do think this isn't the problem,

this is the problem I'm facing.

but eitherway, if you guys have a clue to what am I supposed to do, I'm gladly appreciate ;)

[kutilanak]

I've finally fixed the problem. It's about how you make the dataset, the tone won't affect a thing.

And when you're training the dataset for roads like me, don't forget to turn off the horizontal flip.

I'm closing this