Export training data for deep learning

11-05-2019 03:00 AM
New Contributor II

Hi all!

I'm trying to learn about python and deep learning. I have tried to follow the sample notebooks and watch the training courses but now i'm stucked.

I have done some training samples in Arcgis Pro with the training samples manager and saved it as a shapefile.

Then i export the samples with GP-tool "Export Training Data For Deep Learning" with KITTI Labels as Meta Data Format.

In jupyter notebook i write this to load the samples

from arcgis.learn import prepare_data
data = prepare_data('Path to my samples')

The error i get

~\AppData\Local\ESRI\conda\envs\arcgispro-py3-clone4\lib\site-packages\arcgis\learn\_data.py in prepare_data(path, class_mapping, chip_size, val_split_pct, batch_size, transforms, collate_fn, seed, dataset_type, resize_to, **kwargs)

572 return ner_prepare_data(dataset_type=dataset_type, path=path, class_mapping=class_mapping, val_split_pct=val_split_pct)

573 else:

--> 574 raise NotImplementedError('Unknown dataset_type="{}".'.format(dataset_type))


576 if dataset_type == 'RCNN_Masks':

NotImplementedError: Unknown dataset_type="KITTI_rectangles".

What am i doing wrong?

Regards Mattias

0 Kudos
12 Replies
by Anonymous User
Not applicable

This is helpful, but I am still wondering why the KTTI Labels export is not working - could you elaborate on why the PASCAL method works, but KTTI labels does not? Thanks! 

0 Kudos
New Contributor

I am also curious about this issue with the KITTI labels. I have the same error as Mattias, but I can't switch to PASCAL since I want to use the training data for pixel classification. Any ideas? My KITTI labels already have points, so that can't fix it either... Thanks!

0 Kudos
Esri Contributor

Hi Wimala,

If you want to do pixel classification you can use the format 'Classified Tiles'.



0 Kudos