AntoineH/smart_augmentation
1
0
Fork 0
mirror of https://github.com/AntoineHX/smart_augmentation.git synced 2025-05-03 11:40:46 +02:00
Code Issues Projects Releases Packages Wiki Activity

Rapport + readme

Browse source
This commit is contained in:
Harle, Antoine (Contracteur) 2020-03-26 11:24:05 -04:00
parent 2ec5892c42
commit 55d73fab62
3 changed files with 4 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

0
augmentations_randaugment.py → Old/augmentations_randaugment.py
View file

6
README.md
View file

@ -69,7 +69,7 @@ The following Tensor based transformations are available:
..* 'Posterize'
..* 'Solarize'
A side effect of these Tensor based transformations is that they will outperform, in terms of speed on GPU, their non-PyTorch version as the sized of the images batches increase.
A side effect of these Tensor based transformations is that they should outperform, in terms of speed on GPU, their non-PyTorch version as the size of the images batches increase.
**Please note** that each has defined fixed bound for it's magnitude parameters in 'transformtions.py'. This values are inspired from 'AutoAugment' but it could be needed to change those (especially for images size different from 32x32 pixels). See below the default values used :
@ -87,7 +87,9 @@ Transformation | Min | Max
'Posterize' | 4 | 8
'Solarize' | 1 | 256
Those parameters needs to be float value to allow differentiable transformations and can be configured in JSON files format (see config folder).
Those parameters needs to be float value to allow differentiable transformations and can be configured in JSON files format (see 'config' folder).
It is recommended to use the 'invScale_wide_tf_config.json' config.
There's also other transformations available in 'transformations.py' but they're mostly intended to test purposes and not recommended in general.

BIN
docs/Automatic_Data_Augmentation.pdf Normal file
View file

Binary file not shown.