Open Sauce Fun
Its a Nazli, Its a Crab, Its an Indian Actor.
Created on 25th September 2022
•
Open Sauce Fun
Its a Nazli, Its a Crab, Its an Indian Actor.
The problem Open Sauce Fun solves
For SSoC'22, I had the opportunity to work on following Stuff:
For Deep Learning Simplified Repository:
-
GrapeVine Leaves Identification: The Project aims to provide successful Identification of 5 different types of Grape Leaves. (Ak, Ala idris, Buzgulu, Dimnit, Nazli). The best Model achieved an accuracy score of 80%.
PR: #129 -
Sea Animals Detections is a simple Deep Learning Model Analysis Notebook that compares 3 Basic DL models and the accuracies they have upon training.
In today's world, automatically classifying the data can save countless human hours of manually labeling images. This particular model can successfully classify b/w 19 different Sea Animals whose photos have been taken in their natural habitat. The best Model achieved an accuracy score of 81.7%.
PR: #132 -
Indian Actors Recognition: A more advanced project that aims to classify 135 different Indian Actors. The model still achieved an astonishing 66% Accuracy for this project. The best model achieved an accuracy score of 66.6%.
PR: #138
For PortfolioShop Repository:
-
I worked to fix the boilerplate code of the manifest file and filled it with related content. Also generating better favicons for maximum support in browsers.
PR: #138 -
After various research into most compatible meta tags for best cross-platform support, I found few tags that will result in same and added more favicons sizes including fixes for Apple Touch Icon which expects a 180x180 icon with name apple-touch-icon on many devices.
PR: #148
Challenges I ran into
For Deep Learning Projects:
Due to my lack of knowledge of Image Feature Extraction & Pre-processing, a lot of crucial data was missed, thus leading to overall lower accuracies in the models.
Working with Image Dataset for the first time, I had issues splitting the data into test & train randomly took some effort of trial and error.
Using a model for FaceNet was the most difficult challenge as often many files were incompatible or were throwing errors, so finding the drive link for the pre_trained weights was helpful.
For PortfolioShop:
Many browsers & OS have different requirements for favicon sizes and thus a lot of confusion was there. Thanks to some nice Stackoverflow explanation, I got favicons in most popular sizes allowing for maximum support.
Also, Meta Tags can bloat in size very soon when you try to have every possible sizes, so finding appropriate meta tags that can have most impact on user expereince was rabbit-hole. Thanks to realfavicongenerator, I was able to swiftly find best suited meta tags to work best for the project.