__WORKFLOW_CODES.txt

#AHVAZ PALM TREE SEGMENTATION BY MOIEN RANGZAN

The final processed data sits at: 
E:\_palm\dataset_v2\patched_has_tree\w_palm

the data that was uploaded to kaggle sits at:
E:\_palm\__final_keras_imgen_fomat\data

### dataset_v1:
is just images i uploaded to apeer.com to mask

### dataset_v2:
after masking palm trees in apeer.com I download the results in _palm/mask in two folders "palm" which is only normal palms 
and "palmandcpalm" which we have both palms

background:0
palm:1
cpalm:2

I copy and pased "palmandcpalm" into _palm/dataset_v2/mask

then renamed iamges from "EP-11-11323_0190_0003" to 0003 using "1_rename.py"

then used "2_make_masks_binary.py" to save these mask which are .tiff and 012 as .jpg with 0 as non-palm and 255 as palm pixels.



3- patched
converet images to pathces using 3 "3_patchify_images.py" and "3_patchify_mask.py"

4-has_tree "patched_has_tree"
devided images that has palm in them and the ones that dont into two folders

5-pick test images
moved 1 in every 10 images into test folder for validation of neural network.

6- used augmented sreenicode (which had problem that i resolved)
to created augmented images

7- remove_low_ratio.py
some of the augmentet images has a really small portion of a palm in there, we dont want that!
so we remove images adn their corrisponding mask if the ratio of 255 in a mask to 0s is les than 1 persent


###NOTICE 
since I saved all the files as jpg theres a prboblem with its coperesion, which creates values between 0 and 255 (next time avoid this and use png)
so make sure to use thresholding to resolve this berfore NN taringing.

8- add empty
ading some empty images so the nework learns there coulde ber scense that there are not palm trees in them.