CS 194-26: Pre-canned Final Project

Kin Seng Chau - cs194-26-aae

Part 1: Vertigo Shot

Overview

A sequence of pictures of the same object are taken with different focal lengths from different distance from it. We can observe the "Vertigo Shot" effect caused by the changing field of view (FOV) of subsequent pictures.

The fan heater sequence is taken in my room with my camera placed on the ground. It is pretty easy to capture this sequence of pictures, since I just have to move my camera backward and forward and zooming.

For the table sequence, without a tripod, it is not as easy to take a sequence of pictures which looks natural. The pictures shown below come from my fourth trial. I held the camera and backed off from the table as I zoomed into the pictures. The major challenge was to keep the MacBook screen size consistent and centered as I moved with my camera.

Camera: Sony NEX-3N

Lens: Sony SELP1650 16-50mm f/3.5-5.6 OSS Lens

Fan Heater

The fan heater that supports my motivation during the RRR and finals week.

f = 16mm f = 18mm f = 22mm
f = 28mm f = 35mm f = 50mm
gif

Table with CS194-26

Messy table with the homepage of one of my favorite classes at Cal

f = 50mm f = 44mm f = 37mm
f = 32mm f = 26mm f = 22mm
f = 16mm gif

Part 2: Focus on Miniatures

Overview

In this part of the project, we create fake miniatures by selecting a focus plain, masking it, and applying gaussian filter to the region outside of the focus plain. We got several sets of images below demonstrating the effect of this simple trick.

This project is pretty fun to work on, specially the time I spent on looking for the footage I recorded a while ago. Also, instead of defining a focus line, a mask has to be drawn by the user either using the code provided or using software like PhotoShop. It turns out, if the region is selected properly - one way is to have the boundary of the mask align with edges in the picture, the results appear to be pretty natural and satisfying. Last but not least, the sigma value for the Gaussian filter matters. Depending on the source image, a too large sigma value would result in a very unnatural transition between the two regions, where you can observe obvious artifact.

Special thanks to Nikhil Uday Shinde for the masking code from project 3.

Source images

source mask processed
source mask processed
source mask processed

Internet images

Hong Kong

source mask processed

JR

source mask processed

My images

Macau - my hometown

source mask processed

Rally at the Big Game 2018

source mask processed

Countryside in Kyushu, Japan

source mask processed

Tools in the Embedded Systems Lab (EECS149 class)

source mask processed

Note

This is an awesome class. Big thanks to professor Alexei Efros and the GSI Taesung Park and Shiry Ginosar for making this class so enjoyable.