Grasp train: Image Resizing

[ahundt]

We need to be able to resize images and the corresponding image coordinates to be able to train on the large dataset in a reasonable amount of time, since current training time per epoch is >12 hours. It is also important to update any image coordinate features accordingly, and ensure

• RGB uses a high quality interpolation
• image coordinates are resized accordingly
• implement get_training_tensors() for surface relative poses with coordinate labels
• Other optional TODO items have been moved to grasp dataset Depth resizing, coordinate extents #364

#362 Basic rgb image resizing is implemented as of 00ef348:

#362 image color augmentation is working as of 580186e:

However, we still need to implement the surface relative training tensors that incorporate this data.

This issue is part of the goal outlined in #296, surface relative training.

[ahundt]

This item is complete, just need #362 to be merged