Milestone 4 Presentation

Task Matrix for Current Milestone

Task 1 -Improve accuracy and decrease runtime of CNN

• Improved CNN model significantly by changing it from binary to multi-label classification
• Done fine-tuning of VGG16 layers
• Tuned hyperparameters and model layers to get much better accuracy
• Current Accuracy:

Training: ~95%

Testing: ~89%%

Task 2 -Improve Data Augmentation Code

-Added code that enables augmentation to include filters which would mimic different weather conditions and time of the day.

Example: shadows ,brightness , Rain

- Improved the data augmentation to generate much random and balanced dataset

​

​

Task 3 - Update Website

​

​

​

​

• About
• Members
• Contact us

Report Page

• Created report page
• Made html <form>
• Followed design shown on right

​

​

​

​

​

​

Task 4-Program to connect CNN

> Currently our neural network prediction is a vector of 4 numbers

> Neural network output is stored as a dictionary where Keys are Zones and Values are Binary indicating whether there are any spots available or not

Task 4-Program to connect CNN

> Javascript code has been written to fetch the JSON object created in the backend on Python

> It takes the JSON object and according to the value for each zone, overlays the appropriate indicative box on the parking lot map

Task 5

Creating a Test Dataset:

• Manually created a dataset for 200 images for testing Neural Network Model.

Milestone 5 Task Matrix

Questions