search

TIBCOSoftware / catalystml

CatalystML is an open source specification for real-time feature processing, purpose built to transform data for machine learning models.
BSD 3-Clause "New" or "Revised" License
2 stars 3 forks source link

keep removeAlpha #20

Closed abramvandergeest closed 5 years ago

abramvandergeest commented 5 years ago

bram 11:32 AM so random q when doing image processing I have two operations, removeAlpha and img2tensor I am trying to decide whether to have removeAlpha operate on and image type or on a matrix/tensor when I realized that if you are removing the alpha you are probably already converting to a tensor (this is how I did it for the hands on) so should I just put it in as a flag in img2tensor and drop removeAlpha? mellis:flynn: 11:34 AM maybe img type in the event someone is using something diff than tf, etc? (edited) bram 11:35 AM tensor in this case is the mathematical definiton meaning a multidimensional matrix and not the TF type which all ML stuff would use the multi-D matrix mellis:flynn: 11:36 AM still tend to think that it should stay as its own operation… gives you more flexibility… i think bram 11:37 AM I am not sure image types can hold alpha less images? am I right - also I am looking to see if I have a remove operation that would be equivalent elsewhere mellis:flynn: 11:38 AM ahh if thats the case then i guess no need to have its own op :slightly_smiling_face: bram 11:39 AM I don’t think we have a remove specific row in a random dimension operation, but I still think we are good

abramvandergeest commented 5 years ago

@mellistibco - so we agreed to move removeAlpha into img2tensor as a flag

abramvandergeest commented 5 years ago

So after more discussion we will leave removeAlpha as its own operation such that it inputs and outputs an image type.