Devlog 9

Buttons for Demo

Finally completed the button designs for the demo part of the website when I show what augmentations I use for every model. Added hover animations and a (I assume) novel way of tracking the active state of buttons on Javascript. It uses an active class and assigns it to the clicked button by removing it from all others, ensuring that 2 buttons of the same class are not selected together.
I have also made the website=>
This is a test website and is subject to changes

p.s. If the link seems broken check the project page!

Devlog 8

Section 1 and Pixelated Borders

Created and wrote the first section with the problem statement and a short description of the project and my motivations. I also added the pixelated corners of my website to continue with the pixel art theme. I am going to start working on the augmentation tester and model architecture next.

Devlog 7

Background Star Twinkle and Base Sections.

I changed the JavaScript section to have random timings, size and positions of the stars in the script. I also changed the css to reflect the same and also added the transform command to the keyframe. I also set up the font (pixl sans) and also made the headers and base divs for the sections.

Devlog 6

Small explanatory website.

Working on making an explanation for this project to help in shipping and also to show to people. I have started working on the base HTML and CSS, but I have also added small twinkling stars in the background.

Devlog 5

Early stopping + Test Function

Added early stopping to prevent the model from overfitting and to preserve the best values. Added scikit-learn to work on the testing side and to help with storing the data for the multiple runs I do. The be-all and end-all will be running the model multiple times with different augment values and to see the sweet spot where the augmentation of the dataset starts to help generate more accurate results.

Devlog 4

Added Augment Number and Seed, Added Validation Test Set and introduced CSV files for proper attribution.

In Augment.py I finally bit the bullet and changed the script to work with titles like the name of the file. It will now also generate reproducible random results based on a seed function and number augments needed. In the Main.py I added validation data step to more clearly see where the model was over fitting. Also added the 2 data CSV files.

Devlog 3

Ok, so I just finished with the testing data set and logging all the data in the csv files, I hope to automate this in the future, but currently I’m working on making the data augmentation pipeline faster.

Added Validation and History for Tracking.

I removed the dataset from git status updates and will update it to another website or upload it git in the last commit. I added a validation dataset (splitting) and added it to model.fit. On top of that to track the model, I added history and charts to track the model.

Devlog 1

Augmented the Data.

Fixed file structure again, making it more cleaner and using .gitignore properly. Fixed the augment.py to work with the entire file at once instead of needing to select each file seperately. Fixed the Readme file to have a format, will update it as and when I complete said project.

To-do

• Input changes to the CNN for effective learning
• Adding a test/train split
• Documenting the changes that occur when Batch Size/Epochs are changed. (not really sure on this one)