June 28 - June 29
Our AI Camp batch was focused on Computer Vision. In the first couple days, each of our six team members focused on presenting project ideas, so we could decide by voting. Some of our most voted ideas included the Soil Moisture Detection and the Sign Language Translator, but we voted unanimously for the Fruit Ripeness Detection, which we later named Fripen.
June 30 - July 6
After finalizing our project we split into three groups of two people each and assigned a fruit to each group. Within each group, one person collected the ripe images of the fruit and the other did unripe images. We used SerpAPI to scrape the data.
July 6 - July 7
Since most of the people had no exposure to machine learning before this camp, our instructor gave us a lecture about machine learning and everything we need to know about it. This included the types of machine learning projects, Neural Networks, and some basic features of Neural Networks.
July 7 - July 11
Members from our Data Science team used Colab to train 8 models with different versions of our dataset and different parameters to find the best model. Then, they used Wandb to analyze and compare the performance of each model and finalized the model that performed best.
July 11 - July 14
After training our model, we created a website using HTML, CSS, and Bootstrap. On our website, we created a home page, about us page, data page, results page, and the journey page.
July 14
Once our website was fully functional we just had to deploy our model to the home page. This simple task marked the end of our project!
All Day Any Day!
We faced a lot of different obstacles and challenges while working on this project, some being small hiccups along the way, while others big enough to potentially slow the project down. One of the biggest obstacles was getting the perfect model to deploy on the website. We kept training models, trying to improve on the last one, and we had to train nine models before choosing the ninth model to deploy. enough data for the model. After training some models, we felt that bananas and apples were underrepresented in the model, and had to annotate more images in Roboflow so that the model could recognize them with more confidence. In Web Development, there were many small formatting problems that had to be solved, as well as many large resize-friendly issues while finalizing the website.