Fix inverted inference results because of issues with the config files and channels while running inference

[romeokienzler]

No description provided.

[romeokienzler]

Replicate with Michal - show to @Joao-L-S-Almeida and (@romeokienzler @biancazadrozny)

[romeokienzler]

@Michal-Muszynski can you please tell us how to replicate?

[romeokienzler]

Feedback from @Michal-Muszynski :

When we ran inference with specified input channels (following terratorch guidance), the results were inverted, but when I removed this channel part --data.init_args.predict_dataset_bands [1,2,3,4,5,6] from my command line ..., the predictions were not inverted.

https://ibm-ai-apps.slack.com/archives/C07BSEY2LAH/p1723046232648629

[romeokienzler]

@Michal-Muszynski can we close?

[Michal-Muszynski]

@romeokienzler I have tested the updated terratorch inference in the context of inverted results. Now, it seems that predictions are not inverted for almost all validation tiles.

[Michal-Muszynski]

Feedback from @Michal-Muszynski :

When we ran inference with specified input channels (following terratorch guidance), the results were inverted, but when I removed this channel part --data.init_args.predict_dataset_bands [1,2,3,4,5,6] from my command line ..., the predictions were not inverted.

https://ibm-ai-apps.slack.com/archives/C07BSEY2LAH/p1723046232648629

That was my description of the inverted inference