philip/image_depixelation
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

update readme

Browse Source
This commit is contained in:
Philip Meixner
2023-12-09 17:52:35 +01:00
parent c124e40a20
commit 7414ad5d51
1 changed files with 8 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
README.md
View File

@@ -1,7 +1,12 @@
# Image depixelation project
Training a model which's task is to recreate a pixelated area of an Image
In this repository you can find a project I did in university last semester, where the aim was to train a model which can recreate a certain area of an image which was pixelated before being fed into the model.
In this repository you can find a project I did in university in the summer term 2022, where the aim was to train a model which can recreate a certain area of an image which was pixelated before being fed into the model.
![Image of the task](task.png "Task")
The dataset which was used to train the model consisted of 100 images per student, which added up to about 35 000 images.
Using image augmentation I doubled the amount of images using a little script, which can be found in "image_augmentation.py". Preprocessing of the images was handled by the "prep_image.py", "stack_with_padding.py" and "create_dataset.py" were used to bring the images into datasets which were suitable for PyTorch Dataloaders.
In the file "model.py" you can find my final model architecture and some prior attempts. The final model architecture is definetely simpler than the others, simply because the more complex models did not perform well due to bad generalization.