Week 3: Building my first CNN
Anshul B -
Hello and welcome back to my blog! My name is Anshul Baddi, and I’m excited to share the latest updates on my research project. Over the past week, I have deepened my understanding of how the CNN works and begun researching the Depth-from-Defocus (DFD) method and monocular depth estimation deep learning models (MDE). The components of our code include the CNN to classify and identify objects, the DFD to calculate short distances, and the MDE to estimate long distances (greater than 3 feet). While most of my teammates are currently focused on learning the mathematical foundations of machine learning, I have been independently working on developing my own CNN.
Advancing CNN Development
One of my primary tasks this past week has been creating my own Convolutional Neural Network (CNN). Using Jupyter Notebook, I am designing and training this neural network on a dataset containing images from classrooms and offices. This approach has allowed me to refine my understanding of the CNN architecture and gain practical, hands-on experience with building a machine learning model.
Researching Depth-from-Defocus (DFD) and Monocular Depth Estimation
In addition to CNN development, I have started researching DFD and monocular depth estimation techniques. These methods will play a crucial role in ICON’s ability to perceive depth and calculate distances from obstacles using a single camera. Understanding how these models work will help in developing the next phase of our system.
What’s Next?
In the coming weeks, I will continue refining my CNN model and integrating it with the hardware. Once I achieve satisfactory accuracy with the image classification, I will shift my focus to implementing DFD-based distance estimation. Thank you for reading this blog, and stay tuned for more updates on ICON. I have attached a snippet of my CNN below!
Comments:
All viewpoints are welcome but profane, threatening, disrespectful, or harassing comments will not be tolerated and are subject to moderation up to, and including, full deletion.