Trevor Russo, Week #6: Putting a (Binary) Mask over CMU

Awful title aside, this post, which includes last week and all progress up until today, is about my breaking free from the cycle of failure that had been 3D Modeling. Last week was a little difficult, as I spent the first few days as usual just plugging away at the modeling software. After a while I became frustrated beyond belief, and was about to lose faith in my ability to finish the project when my fellow intern, Joe, told me about someone that had also worked on 3D Modeling, named Pedro. I texted Pedro and asked if we could meet up to try out the software that he used, Agisoft Photoscan. He came down to the lab and gave me a quick tutorial, but the real breakthrough came when I showed him the Berkley Dataset I had been using. He noted that the reason the pictures weren't working was because the turntable they had been positioned on was static but the object was not. When the software, Autodesk Remake for example, attempts to reconstruct the scene, it automatically estimates the camera position and stitches together the photos based on this information. Because the actual turntable was static the entire time, the algorithm was confused. I wish this information had occurred to me long before I lost 3 weeks to faulty modeling, but I took this as a victory in itself.

Then I remembered what Venkat had told me about a possible solution, and for the first time, I decided to mask my photos. Masking is where a binary mask, which is a collection of 1s and 0s, is applied to a photo, essentially "turning on or off" the pixel in that location. After applying the mask, the photo eliminated the background. And when further plugging these new images into Autodesk Remake, it produced a solid model of the Cheez-It Box pictured below. Now the Berkeley pictures came with binary masks, and the photos that I and others will be producing won't have them, but I have been assured by Venkat that the program that will produce masks will be written separately after I leave.

 Before Masking

After Masking

  

Between that, I decided to try and take my own photos, creating a dataset for my water bottle. Unfortunately the model was lacking, and by lacking I mean lacking its middle portion. And with 2 days to go, it seems like trying again doesn't seem like the best use of my time. So, in order of priority, my tasks follow as such:

1. Write a Program that AutoWhiteBalances the pictures

2. Produce more models from the Berkley set and using my masking program mask them

3. Test out my Photo program with an actual kinect

4. Write a report so that others who use my software know what to actually do.

And that's about it. I'll write another post on Saturday on the way home, but after 6 wonderful weeks, it's somewhat surprising and saddening to know that I'm leaving a place that has been like a second home to me. I love everything about the city and the college, and hopefully I can continue here for four more years.

Other Photos:

The Water Bottle I took photos of

 A UAV about to take flight towards a goal

A King upon his Throne