suggestion/feedback on image display

Discuss your experiences with and ideas about Stardust@home here.

Moderators: Stardust@home Team, DustMods

Post Reply
Chuck Crisler
Posts: 30
Joined: Wed Aug 02, 2006 9:44 am
Location: Windham, NH

suggestion/feedback on image display

Post by Chuck Crisler »

I am happy that I have had a chance to participate in this project. However, there is something that annoys me about the VM that could be improved. The images are displayed from top down, but the important images are on the bottom. The implementation, coupled with the way XP randomly decides to re-arrange the page file and network glitches (on virtually every movie) means that I spend 3-5 times as much time looking at images above the surface as below. If you displayed from bottom up, then I would spend more time (or make my decision faster) looking at the truely important sub-surface images. That would mean that I could process movies faster and your servers would retrieve fewer unnecessary images (I define images above the surface as generally unnecessary). However I am still convinced that you could easily automate this process. The advantage of automation is that, if you learned a new pattern, the pattern recognition profile could be changed and all of the images re-scanned. Nevertheless, I am still having fun!

GelDelve

Re: suggestion/feedback on image display

Post by GelDelve »

Chuck Crisler wrote:The images are displayed from top down, but the important images are on the bottom. . . . If you displayed from bottom up, then I would spend more time (or make my decision faster) looking at the truely important sub-surface images. That would mean that I could process movies faster and your servers would retrieve fewer unnecessary images (I define images above the surface as generally unnecessary).
Yes, I've often thought the same thing. I'm assuming that you are proposing that the order of the frames be reversed, that is the bottom-most frame corresponding to the lowest focus bar would load first then each frame above would be loaded in that order. I'm glad you brought the subject up. It seems like that could be accomplished without much effort, but I don't know enough about programming to really judge it.

There may be some users who would be resistant to such a change and there would have to be some kind of cut over and learning curve. CerealKiller wrote a computer program that only loaded the first and last frame of each movie, so it would seem that there might be a way to load the movie frames bottom up by the users choice rather than UC Berkeley making any change at all. Maybe someone in the forum knows if it's possible and how to do it. I'd be much interested in that.

jsmaje
Posts: 616
Joined: Tue Aug 15, 2006 8:39 am
Location: Manchester UK

Re: suggestion/feedback on image display

Post by jsmaje »

Chuck Crisler wrote:The images are displayed from top down, but the important images are on the bottom. . . . If you displayed from bottom up, then I would spend more time (or make my decision faster) looking at the truely important sub-surface images.
I also thought that at first, and routinely started from the bottom frame up (once all were loaded), but have subsequently reversed my procedure and always now start from the top. The crucial thing is to identify the surface and understand its topography, e.g. how it might be slanted, distorted, grooved, debris-strewn, etc. Only then can you be sure that in any section of a movie you are looking at something genuinely below the surface, and whether it may just be an inclusion or possible track. After all, the aerogel recovered from this mission is far from perfectly flat or uncontaminated!

Too many pictures have been posted of so-called 'tracks' that are nothing of the sort - just out-of-focus debris (giving a ring-like appearance), but are merely on the surface. This is bound to happen when the microscope focussing has been too high initially and not even penetrated the actual surface, or because there is a valley in the aerogel, etc.

That's one reason why I was sceptical about CerealKiller (looking at just the first and last frames) - experience has shown that a significant proportion of movies are not well-focussed, nor could possibly be expected to be well-focussed because of a convoluted surface etc.

I think the scientists here have had a good deal more experience of looking at these things than we have, and that there is good reason in what might otherwise be thought their madness!

fjgiie
DustMod
Posts: 1253
Joined: Sat May 20, 2006 8:47 am
Location: Hampton, SC, US

Automaton

Post by fjgiie »

Hi Chuck Crisler,

May I address the last part of your post about automation?
Chuck Crisler wrote:I am still convinced that you could easily automate this process. The advantage of automation is that, if you learned a new pattern, the pattern recognition profile could be changed and all of the images re-scanned. Nevertheless, I am still having fun!
Here is what I have learned over the past eight months about automation. As far as I know the tiles are photographed by an automatic microscope with a digital imager. Itwill take up to a year at the automatic microscope/digital camera at Houston, TX to photograph all the tiles. May finish by late spring or early summer. I believe there will have been around 700,000 focus movies made by completion of the project.

From here we learn that there are around 4000 dusters that log in and search every day. From my experience, 80 real movies can be viewed and graded in one hour. That would be 160 in a two hour day per duster. 160 real movies X 4000 dusters equals 640,000 focus movies a 2 hour day. If we want the movies looked at 10 times that would be 7,000,000 movies. So 7,000,000 divided by 640,000 movies per day would be eleven days. There is no problem looking at the movies.

The problem is scanning and photography at the automatic microscope at the Curatorial Facility at Johnson Space Center in Houston. There lies the bottleneck. What type of automaton needs to be done there? :)
http://stardustathome.ssl.berkeley.edu/about.php

Thanks,

fjgiie

Chuck Crisler
Posts: 30
Joined: Wed Aug 02, 2006 9:44 am
Location: Windham, NH

Post by Chuck Crisler »

The big advantage of automation is that you can repeat an entire scan of all of the images anytime that you learn sometime. Simply changing the 'rules' allows you to re-examine the data from a different angle. There has been some fascinating work on pattern recognition. There is one system that I know of (which means that there are probably more) that allows a mathematical system to actually learn from mistakes (Bayesan learning). It allows you to recognize enormously complex patterns using a huge number of different dimensions, things people just can't do across endless numbers of images. Note that I am saying that after 'some guy' examined over 240,000 images from Mars and found (and remembered!) 2 images of specific craters that showed changes over 6 years, suggesting water AND significant meteor impacts. Nevertheless, automation can be more reliable. Using a mix of people and automation MIGHT produce better results, though we wouldn't be able to participate! :(

Post Reply