split_raster() function outputs are distorted

[MoAbbasid]

The preprocess.split_raster() output crops are distorted.
Im running the same code on the training notebook, Im trying to test finetuning a model for more dense forest imagery to see how it performs.

code:
crop_dir = "/content/train_data_folder" train_annotations= preprocess.split_raster(path_to_raster=YELL_train, annotations_file=annotation_path, base_dir=crop_dir, patch_size=400, patch_overlap=0.05)

I just changed the output dir
Screenshots:

not sure if its related to the way the pngs are cropped or the way they are saved.
any idea why this happens?

Thanks

[bw4sz]

There is no error here. This is what the test data from NEON looks like. Completely unrelated to DeepForest. They georectify using a LIDAR mask that distorts the RGB imagery. Check out the original image, you will see it there as well. Closing.

[MoAbbasid]

@bw4sz, you're right, I looked at the large area and it does indeed look like that,
bare with me these noob questions while I familiarize myself with deepforest