Using Mode instead of median of mean for automation of image prediction

hamidrezaomidvar / LINDER

LINDER (Land use INDexER) is an open-source machine-learning based land use/land cover (LULC) classifier using Sentinel 2 satellite imagery

13 stars 2 forks source link

Using Mode instead of median of mean for automation of image prediction #12

Closed hamidrezaomidvar closed 4 years ago

hamidrezaomidvar commented 4 years ago

Let's keep this in mind, and try it late on to see if we can fix the automation of image pickup problem. Theoretically, this should work if we have many images.

hamidrezaomidvar commented 4 years ago

@sunt05 I would like to spend some time in near future to enhance this package. The first thing I like to do is to use Mode function as we discussed to automate the prediction part. This would help us to speed up the model as we do not need to do the steps after the prediction step for all images. Other enhancements that I have in mind are:

[ ] breaking down some the files to sub-modules
[ ] Make test cases and automate them using GH actions
[ ] Make documentation
[ ] Validation (this is more in the science part)

Please let me know if you have any more suggestions.

sunt05 commented 4 years ago

These all sound good to me except for Azure: we may consider GH actions, which is more tightly integrated with GH (obviously). I used Azure for SuPy because GH actions was not out then.

hamidrezaomidvar commented 4 years ago

These all sound good to me except for Azure: we may consider GH actions, which is more tightly integrated with GH (obviously). I used Azure for SuPy because GH actions was not out then.

I see! then I edit it to GH actions.

hamidrezaomidvar commented 4 years ago

@sunt05 I implemented this. It works very well, now it first predict the images and based on the frequency of prediction for each pixel, it outputs a final prediction tiff file. So the tasks after this are done only over one file not all of them. So only one output for this package now.