This page contains a list of some research oriented personal projects.

Clothing Structure Information from Product Photos

The idea behind this project came to me while browsing back when it launched in 2010: I was increasingly frustrated with its inability to differentiate between tops which were fitted because they were made out of a jersey/knit material and those which were fitted because of construction techniques. Much more recently, after listening to someone describe their approach to online clothes shopping, I began wondering if it would be possible to extract proportion information from the images as well. Thus this project has two parts:

  1. Identify attributes about the clothing. One set of attributes includes the sort of silhouette aspects, such as the sleeve length of a shirt (tank, cap, short, 3/4 length, or long) and the style (ex. straight, pagoda, bell). This set of attributes also includes information such as the neckline style and hem style. The other set of attributes includes the details and construction aspects such as seam lines. This set would also include collar style; number and style of darts; number, style, and location of pockets; and whether a shirt has a yoke or a peplum.

    Identifying these attributes is important for two reasons: 1. personal taste and preference searches; and 2. fit (certain construction styles are more likely to fit certain body shapes better).

  2. Identify proportions from the clothing. Different clothing companies have different base body models; this infographic illustrates the variation between circumference ratios, but variations in height and length are also likely. The objective of this part of the project would be to identify proportions between torso length and circumferences (bust/chest, waist, and hip), and determine if the proportion information gathered could be used to estimate the fit of a clothing item.

    For this part of the project to work, many assumptions and estimates will need to be made, such as the accuracy of product photos in representing the clothing, and the types of grading methods used and the effect those have on the torso length relative to width. I am not sure that these assumptions will hold true, or that the estimates will be generalizable.

User State/Action Tracking via a Webcam

When knitting or crochetting from a pattern, I often forget to mark where I am in the pattern when I set the project down, and then must spend some time locating my position when I again pick up the project. The idea here is to use a webcam to track stitch motions and automatically update the current location in the pattern. Eventually I hope to be able to integrate this with a stitch-charting webapp I am developing.