 Stephen
|
20167ca337
CHanged all filenames to be parsed with os.path so that they should work on windows. Testing required.
|
2 years ago |
|
Stephen
|
2a0275817b
Updated the cardinality line to modern code.
|
2 years ago |
|
Stephen
|
a668338804
Fixed the loader bug. The tf.data.experimental.cardinality() function returns a tensor, not a number. So I converted it to a number with numpy. Note the experimental tag is no longer required.
|
2 years ago |
|
Stephen
|
92a366f658
BROKEN. Made the val-test splitter accept variables instead of hardcode. But I can't fix the bug yet.
|
2 years ago |
|
Stephen
|
2aac5d9308
Updated a folder name to be consistent. Added a tutorial.
|
2 years ago |
|
Stephen
|
c829284979
Added the validation set back in. I forgot that model.fit() can only automatically split a dataset into train/validation automatically if the input is not of the tf.whatever.dataset type. Which is dumb.
|
2 years ago |
|
Stephen
|
f84119cbc7
Minor edits. Updated README.
|
2 years ago |
|
Stephen
|
e2d9b9e575
Made creating folders simpler. Made folder for saving augmented dataset as pngs.
|
2 years ago |
|
Stephen
|
b3998e930b
Created a funcion to save pictures to a file.
|
2 years ago |
|
Stephen
|
1e698ddd99
Changed the name of validation set to testing set.
|
2 years ago |
|
Stephen
|
0f6d096238
Removed training step from this project.
|
2 years ago |
|
Stephen
|
5452034cbf
Removed creation of test set.
|
2 years ago |
|
Stephen
|
04f3c7aad6
Reverted to grayscale.
|
2 years ago |
|
Stephen
|
f8e9180fb5
Janky working version. Increase epochs and omit steps per epoch to get better results.
|
2 years ago |
|
Stephen
|
3042281cf3
Working version with good accuracy in training. Need to make test set evaluation now.
|
2 years ago |
|
Stephen
|
94b1e0155b
CHanged imagesize to a hardcoded 150 for the time being. Working version of data import on train script.
|
2 years ago |
|
Stephen
|
03451ac13c
Changed input to rgb because most image models use rgb as input, can convert to grayscale before use if needed. Downside is the filesize is three times bigger. Added a script for training.
|
2 years ago |
|
Stephen
|
a42423fa4b
Reduced image size back to 100 for now.
|
2 years ago |
|
Stephen
|
7600aedc13
Added printing of the augmented data so we can see it. Added minor upgrades to parallelization.
|
2 years ago |
|
Stephen
|
4b980bfcc5
Found how to get the images out of the prefetchdataset or whatever. Unclean running version.
|
2 years ago |
|
Stephen
|
d9bfcfcf6c
Updated readme; The loader function shuffles automatically.
|
2 years ago |
|
Stephen
|
1c08195815
Added tqdm for nice loading bar animations. Added a helper function for printing. I don't know how to print from the final datasets; they are of a different type.
|
2 years ago |
|
Stephen
|
2818995099
CHanged picture input to grayscale. Fixed printing error.
|
2 years ago |
|
Stephen
|
a90148dd4d
Remove old dataset files before running the script-- so tha they don't accumulate and waste space.
|
2 years ago |
|
Stephen
|
a46b62c287
Enabled saving the dataset.
|
2 years ago |
|
Stephen
|
d01407d623
Updated directory name
|
2 years ago |
|
Stephen
|
1e132676f8
Removed un-needed loop in the image copy phase. Fixed a bug that would download the pictures multiple times if muliple triggers were set.
|
2 years ago |
|
Stephen
|
7155b7230f
Fixed preprocessing. I found out that my tensorflow was really old-- causing some of the code to not work. Using tensorflow 2.7 and numpy 1.22.3 fixed it.
|
2 years ago |
|
Stephen
|
319e2f6b02
Added a thing to automatically download if it is the first run. Also download flags from the command line and as a global variable.
|
2 years ago |
|
Stephen
|
308d5f521f
Merge branch 'main' into mychanges
|
2 years ago |
