Selected Projects
Selected Projects
Embedded Files
- Monocular Depth Estimation using a semantically meaningful Loss function
My model uses a fully convolutional network architecture to map the RGB image to a depth map. A fully convolutional net tries to learn representations and make decisions based on local spatial input by learning the optimal filters that minimize the depth map construction error. The reconstruction error is calculated by a semantically meaningful loss function that penalizes structural dissimilarities between the reconstructed depth map and actual depth map. Refer to this GitHub page for details.
2. Obstacle free path generation via Visibility graph and Dijstra's algorithm
2. Obstacle free path generation via Visibility graph and Dijstra's algorithm
The aim of this project was to find an obstacle free path between a source and destination with an added constraint of visiting some nodes at-least once in its traversal. I used the visibility graph method to method to build a connected graph and then used dijstra's algorithm for find the path. Visibility-graph is a graph of inter visible locations, typically for a set of points and obstacles in the Euclidean plane. Each node in the graph represents a point location, and each edge represents a visible connection between them. This project was meant to solve the problem statement of a robotics competition. [Code]
3. Image mosaicing
3. Image mosaicing
Image mosaicing is the process of combining multiple pictures with overlapping field of view to produce larger field of view image. This project implements a tool for helping the user to create an image mosaic. The user can take images using a hand held camera at different angles with some overlapping area, then use this software to generate a larger field of view image of the scene. [Code]
4. Function mapping using a neural network with a single hidden layer
4. Function mapping using a neural network with a single hidden layer
This project attempts to validate the finite sample expressivity of neural networks with a single hidden layer. I trained the shallow(1 hidden layer) neural net with non-linear functions of various types and visualized the results.[Code]
Page updated
Google Sites
Report abuse