New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Create NPZ file for training #99
Conversation
|
||
Args: | ||
npz_list: list of NPZ files to combine | ||
resize_ratios: ratio used to resize each NPZ if data is of different resolutions |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Is this meant to be a list or an array? does 1 indicate no scaling required?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yes, this is a list that matches the NPZ files being specified, where 1 indicates no resizing required
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
There should be a threshold for scaling. (e.g. 1.01 does not need to re-scaled)
@ngreenwald I think we need to take another look at this PR. It seems somewhat fragile as presently constructed and heavily tailored to static image set. There also appears to be several locations in the code that call for padding on top of existing padding, is that by design? |
For sure, I actually have no idea what the NPZ file for training a tracking network looks like, so I'm guessing as is this won't work for that. Happy to get advice on how to make it more generalizable, or if we should explicitly restrict to static? Yes, the two paddings accomplish different goals. The first padding is to make the full image size an even multiple of the desired crop size for the network. This is because the deepcell_toolbox.tile_image function will index from the end of the image to make up for any non-even division, which isn't the desired behavior since it will override the user-specified stride ratio. The second padding is for data augmentation: it optionally adds blank padding equal to half of a crop size on either side of the image to squeeze an extra crop out of each row/col of the image. |
Hey @MekWarrior, what do we need to do to merge this in? |
Basic functionality for build.py to create an NPZ with specified overlap and crop size.