danpovey
commented
6 years ago That seems to make sense to me. There might be a few utility functions that need to be switched over. I hope Yiwen can take care of this while updating the dsb2018 setup to use more of the core tools.
On Wed, May 16, 2018 at 1:06 AM, Desh Raj notifications@github.com wrote:
@desh2608 commented on this pull request.
In scripts/waldo/data_types.py https://github.com/waldo-seg/waldo/pull/29#discussion_r188504039:
"""This function validates that 'c' is a valid configuration object of type CoreConfig. """ assert isinstance(c, CoreConfig)
- c.validate()
- c.validate(train_image_size)
def validate_image_with_mask(x, c):
@hhadian https://github.com/hhadian yes, Numpy and Scipy use the row-major format, while image libraries such as Pillow or Matplotlib use the column-major format. I think since we are storing all image data as numpy arrays anyway, we should use the (height, width, num_colors) format, which would avoid the need for axes transformations while converting.
— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub https://github.com/waldo-seg/waldo/pull/29#discussion_r188504039, or mute the thread https://github.com/notifications/unsubscribe-auth/ADJVuwQoC1gKBHfo9xhAuVcZe1X-tPAiks5ty7PAgaJpZM4T_I8d .
YiwenShaoStephen
hhadian
desh2608
Removed train_image_size from CoreConfig and added as argument for functions in data_manipulation and validate() for CoreConfig. Also added it as command line argument in training script for dsb2018 with default value None.
N.B: train_image_size is directly used without a None check in several places. I think we should provide a default value for it other than None otherwise there may be issues down the line. Please provide your thoughts.