EnsembleFill: Using Machine Learning to Assist Artists in Image Repair


Help Improve Image Patching/Hole Filling Techniques!

My name is Lesley Istead and I'm a PhD student under Dr. Craig S. Kaplan in the Computer Science Department at the University of Waterloo.

Study Purpose and Procedures

I'm looking for volunteers to help produce/classify images as part of a project to understand and improve image patching/hole filling techniques. This study has two goals: to collect sample holes (things that have been removed from an image), and determine which image patching/hole filling technique produces the best result. The collected responses will be used to produce an ensemble algorithm, which intelligently selects which patching technique is most appropriate for the provided hole and image combination.


This study has two parts. Both parts can be completed online using a web browswer (Chrome or Firefox).


The first part of this study is to collect a set of holes---a single shape that represents a region of an image that requires filling/repairing. The second part of this study is a survey that asks participants to choose the best image from a set of nearly-identical repaired images. Each task or survey should take no more than 3 minutes. The data collected from these tasks and surveys will be used to train and validate a computer model.

Task Details

Part One of this study asks you to complete one of two randomly selected tasks. These tasks are completed online using a web browser (Chrome or Firefox) and require the use of a mouse. Both tasks ask you to draw a hole, a single shape, in the designated area. This hole should represent something you would remove from the image such as power lines, cracks, dirt, coffee stains, etc. Each task should take no more than 3 minutes to complete. No artistic skills/abilities are required. No personal information of any kind is collected during these tasks. The image that you produce will be saved and may be shared with other researchers in the field.


Part Two of this study is a survey. Each "question" in the survey will present you with a set of nearly-identical images. A region in these images has been repaired. You are asked to choose which of the images looks best, or, indicate that there is no difference between them. You can choose how many images you'd like classify, (between 10 and 40). Each "question" should take no more than 30 seconds to respond. No personal information of any kind is collected during this task.

Potential Benefits and Risks

By participating in this study you will be helping improve image patching/hole filling techniques.
There are no risks associated with completing the tasks or surveys in this study.

Participation and Time Commitment

Participation in this study is voluntary. You can complete these tasks/surveys as many times as you wish. Each task or survey should take no more than 3 minutes. If at any time you feel the need to stop your task, navigate away from the page. Partially completed tasks will be discarded.

Confidentiality and Questions

This survey is anonymous in that we do not ask for your name or any identifying information. Once you have submitted your responses it is not possible to withdraw your consent to participate as we have no way of knowing which responses are yours. All data will be stored on an encrypted laptop for a minimum of 5 years. Data related to your participation (survey responses and images) will also be published online (on the personal research page of Lesley Istead). This data will be completely anonymized since we do not collect any identifying information at any point. This process is integral to the research process as it allows other researchers to verify results, avoid duplicating research, and expand on our method. Other individuals may access this data by downloading it as a ZIP file.

When information is transmitted over the internet confidentiality cannot be guaranteed. University of Waterloo practices are to turn off functions that collect machine identifiers such as IP addresses. The host of the system collecting the data is the University of Waterloo. If you prefer not to submit your task/survey responses through this host, please do not sign up for this study as there is no alternative method to collect the data we require.

If you have an questions about participation, or the goals of the study, please contact Lesley Istead at lanortha(AT)uwaterloo.ca

I would like to assure you that this study has been reviewed and received ethics clearance through a University of Waterloo Research Ethics Committee. However, the final decision about participation is yours. If you have any comments or concerns resulting from your participation in this study, please contact Dr. Maureen Nummelin, the Director, Office of Research Ethics, at 1-519-888-4567, Ext. 36005 or maureen.nummelin@uwaterloo.ca.

Consent

With full knowledge of all foregoing, I agree, of my own free will, to participate in this study.

I agree to participate. Click here to begin!

If you do not wish to participate, please click below or close your browser. I do not agree to participate.