use polygons to crop images from master tifs, then rotate them to horizontal/vertical does not resize, usues small buffer to increase polygon size

01_data_prep_rasterio.py

Needs QC before use earlier version, crops only

01_data_prep_rasterio.py

crops, resizes, and pads out images using BORDER_REFLECT_101

02_data_crop_concat_20191025.ipynb

crops, resizes, pad using BORDER ***may need debugging

02_1_data_crop_concat_20191027.ipynb

submission 2

03_1_fastai_train_resnet_20191026.ipynb

qc of results

03_1_1_result_qc.ipynb

sharpen blurred images

04_image_reprocessing.ipynb

as 03_1_fastai_train_resnet_20191026.ipynb but with sharpened images slightly worse result

05_1_fastai_train_resnet_20191027.ipynb

augmented to even out class distribution

04_1_image_repro_even_up_class_numbers.ipynb

NB this is run with data_dir/f'train/rotated/clipped/256' and data_dir/f'train/rotated/wrap/256' data paddings

TODO - combine all the augmented and different padded data for a simgle train and test ds

see https://github.com/rymc/n2d/tree/4b888b31bccf092032c8eac6e82feb06d5461e94

installed under ../unsupervised/image_clustering/tensorflow/josephsdavid/N2D/

use venv N2D n2d_clustering-train_valid.ipynb

Can see differences bw train / val and test - particularly unverified

GAN

see https://arxiv.org/pdf/1809.03627.pdf

https://github.com/eriklindernoren/PyTorch-GAN#installation

/mnt/963GB/Data/Python/Code/GAN/PyTorch-GAN/