GT4 Photomosaic Project

  • Thread starter Thread starter ZZII
  • 68 comments
  • 2,349 views
Awesome! 👍

Pretty soon you'll start having better resolution that GT4 itself. I mean, the wheelwells in that photomosaic of my S800 is already better than how it was in the GT4 picture :p
 
very nice job, 3rd, 5th, 8th, 9th, 10th are all mine :)

keep it up, ur definatly a pro at thi, cause i have no idea what your saying, (thats always a good sign) i think
 
Wow, I never thought putting all those pics together would even made a shape of a car. :lol:

That's awesome work. 👍
 
Progress Report #3:

I got through a few more galleries, bringing the total image count to about 1,450 at the moment. I'm still attempting to go in order of response, so for everyone who has posted links to their gallery more recently, thanks for contributing, but it might take me a while to get to it. If anyone has the time and would like to get quickly entered into the collection, you can still send me a .zip file with a bunch of your images.

I spent some time tinkering with the code to create some way of automatically tracking whose images are used where. Since there are 1,600 images total, a simple list would be hard to follow. My first idea was a kind of grid with coordinates for each image and the username of the contributor. You'd have to count over to the image on the actual picture, though. A much clearer way of doing it is to just print the username right on the corner of each picture (not on the real thing, of course - on a separate "key" image). This turned out nicely, so I will go with that method from now on. (The text-based grid can also be posted for anyone who can't load the full-sized images, which are 4+ MB each.)

Of course, the progress report wouldn't be complete without a new picture, so I present Teaser #3 (big photo contributed by JTSnooks):

teaser3_small.jpg


and a zoom-verson from the new image credit key:

teaser3_key_zoom.jpg

(Anyone want to play the where-is-it-on-the-big-picture game? :) Hint: One of Master Yoda's pictures is a good indicator.)

Here's the text version of the image credit key:
http://web.mit.edu/scolton/www/gt4photomosaic/teaser3_key.html

And for the first time, here is a FULL-SIZE (6400x4800) version of this test render (mmmm...bandwidth :cool: ):

http://web.mit.edu/scolton/www/gt4photomosaic/teaser3.jpg (4.48 MB)

FULL-SIZE image credit key:
http://web.mit.edu/scolton/www/gt4photomosaic/teaser3_key.jpg (5.58 MB)

More to come! I should break the 2,000 picture mark tomorrow.
 
kennythebomb
Wow, that particular one really loved the profile shot of my Buick!

Yea! I didn't notice that it actually appears 4 times in the upper left 6x6 corner...it must fit the color up there well. When I get up to a few thousand images, I will turn the repeat check range up so that repeats won't occur that frequently (not that I don't like the Buick pic ;) ).
 
Progress Report #4:

I took a bit of a break over the weekend, but I'm back to work collecting images now. I finally made it through the first page of posts, giving me a total of 2,025 images at the moment. I'm going to keep going through galleries grabbing as many as I can...it might actually be possible to hit the 5,000 mark I was shooting for originally.

I'm starting to get a good idea of what will come out good as a render. Generally, the tighter the shot of the car, the more detail you will see in the render. Using only 40 images across and down makes it hard to get very crisp detail. I might try using more than 40, but the sacrifice will be smaller small images and more repeats. Also, lighter images have been rendering better than darker ones. Color contrast seems to work nicely too.

For this test-render, instead of looking for a picture that works nicely, I picked a picture that I liked and tried to make it work. The original photo was taken by Div is back.

Original:

photo014edit7iq.jpg


Here's the un-touched render:
teaser4.jpg


I knew that since it was a dark image, it would be a bit of an ugly render. Although it looks pretty good from a distance, some of the detail is lost, I think. The other thing that was missing was the pinkish-orange tint to the reflection - it almost looks green. So, I did two things: The easy fix was cropping the image a little tighter on the car itself. This cut out a lot of the black space at the top, which doesn't render well, and adds detail to the car. The hard fix was adding in the color adjustment that I have been putting off since I wrote this code. It adjusts the colors of each little picture slightly (20% weighting, in this case) in the direction of the average color of the picture in that section. The effect is barely noticeable in the small images, but when viewed from a distance it is significantly better. I left this option so it can be adjusted or turned off completely, since it's borderline-cheating (justifiable, though, when you work with relatively small image collections). Anyway, here's the result:

teaser4_color.jpg


big version: http://web.mit.edu/scolton/www/gt4photomosaic/teaser4_large.jpg
image credit key: http://web.mit.edu/scolton/www/gt4photomosaic/teaser4_key.jpg
image credit key, text version: http://web.mit.edu/scolton/www/gt4photomosaic/teaser4_key.html

I probably won't post again until the 3,000 image mark. I'm still going in order of post, now on the second page. Thanks again to everyone who has contributed...if I haven't gotten to yours yet, I will! :)
 
Sweet, my Viper's there! Oh yeah, and that set was from the passenger-side headlight in the top-left corner :D Found it without looking at the big image even! Anyway, I'll see if I can get another zip file of pictures ready for you soon, great work! The second mustang render looks quite good, actually :)
 
JTSnooks
Oh yeah, and that set was from the passenger-side headlight in the top-left corner :D Found it without looking at the big image even!

Bingo, again.


Div is back
Many pictures from me, including the original.

That's an honour, thanks!

My new desktop ;) :
desk.jpg


Thanks for the pics! :)
 
JTSnooks
You planning on doing any more?

Absolutely. I don't want to leave out anyone who contributed, although I might have to take fewer images from the large galleries to keep up with the posts.

I'll also start taking suggestions for things to render. Generally, tight on the car, light (not a night track), and nice color contrast work best. So if anyone has a particular picture they'd like to see, let me know. I also want to do a GT/GTP logo of some sort.
 
As promised, another progress report at the 3,000 image mark (2,987 to be exact):

I ran into a bit of trouble with the program soon after the 2,000 image mark. (I expected I would at some point.) The problem is that the Windows file-open thing that I am using can only handle so many files at one time. Rather than trying to actually solve the program, I just made some minor changes that allow me to load images in more than once. (So, I can load 2,000 at a time. I should never have to do it more than 3 times, so it's not a big nuisance.) Problem solved as far as I'm concerned. ;)

I've been debating going to a higher resolution to get clearer pictures. (Using more than 40 images across and down to give smaller "pixels.") I'm still leaning toward keeping it as is because I wouldn't want to shrink the small images...they are the focus of it all. The more you shrink them, the more you go from a photomosaic toward just a plain mosaic. Plus, with 3,000 images, there are already a lot of repeats in the 1,600 that compose the big image. With higher resolution, this would be even more of a problem. Any input?

Nobody has sent any suggestions for renders, so I've just been rendering the pictures I like. Send me your favorites! It takes me less than 10 minutes to render them! :) I have plenty of web space to host one for everone who has contributed pictures. The best are "tight and light:" daylight tracks and close zoom on the car.

And of course, it wouldn't be a progress report without a new render. I'm not calling them teasers, anymore, since they are full renders with color adjustment. So, I present Test Render #1:

Again, I chose a picture that I knew would present a problem because it violates half of the "tight and light" rule...last time it was a dark picture, this time it's a wide zoom. I liked the photo, though, and I decided to give it a try anyway. Here's the result:

test1_real.jpg
test1_small.jpg

(left) real image, credit to FWA2500, (right) thumbnail of the photomosaic

From far away (or when I make the picture really small), it is a very nice match. That is the most important thing, I think. From close up, the detail on the car is lost, but by the time you are looking at it close up, the small pics should catch your eye more. At least, that's the idea.

Here is the full-sized render:
http://web.mit.edu/scolton/www/gt4photomosaic/test1.jpg

and the key:
http://web.mit.edu/scolton/www/gt4photomosaic/test1_key.jpg

and the text key:
http://web.mit.edu/scolton/www/gt4photomosaic/test1_key.html

Edit: I just noticed that when viewing the photomosaic in I.E., it automatically resizes it to fit the window but it does a really poor job of it. (It uses a deletion resize...just deletes every n-th line of the image.) If you want a nicer-looking version in 1024x768 or a resolution like that, save the image and use a photo editing program to shrink it using a linear or cubic filter.

I hope to be able to collect as many images as possible from everyone who posted contributions up until this point before I post another progress report. That should bring me near 4,000 I think. At least then I will feel caught-up. Until then, thanks again! :)
 
Another great photomosaic 👍

The resolution at the moment seems really good, however if you're not sure you could maybe re-render this the same pic with a different resolution to show us what it'd look like then?

Keep up the good work
 
amp88
Another great photomosaic 👍

The resolution at the moment seems really good, however if you're not sure you could maybe re-render this the same pic with a different resolution to show us what it'd look like then?

Keep up the good work

Thanks! I don't want to change the code yet, but what I can do is simulate what it would look like by breaking the large image up into 4 smaller ones and rendering each of those individually, then splicing them together. I'll give that a try and post a comparison image in a little while.
 
you are truly amazing....i have a good 200 pics if you want i can aim them to you....midnightxsoldier or nismox4xlife , its looking gorgeous though. Keep up the godly work 👍 :) :dopey: :drool:
 
Ok, here it is layed out in full:

The Resolution Dilemma

The previous renders were all done with 40x40 resolution. That means they were composed of 40 rows and 40 columns of small images, 1,600 in total. Each small image was 160x120, giving a total image size of 6400x4800.

Increasing the resolution would give the big images considerably more detail. As an example, I created an 80x80 version of the same image. It is entirely possible to go somewhere in between 40x40 and 80x80 as well, but for convenience, I just did the two extremes. Due to memory restrictions, the total image size cannot go much higher than 6400x4800 (stored as a bitmap in RAM, it eats quite a lot of memory). So to do 80x80, I had to reduce the small images to 80x60. That is the compromise: less detail on the small images.

Here's the comparison (left is 40x40, right is 80x80):

big image:
test1_small.jpg
test1_highres_small.jpg


small image sample:
test1_sample.jpg
test1_hrsample.jpg


and here is the full-size high-res:
http://web.mit.edu/scolton/www/gt4photomosaic/test1_highres.jpg

Low-Res pros:
-more detail on the small pics
-smaller file size (better compression...5-6MB vs. 20MB)
-fewer repeats
-more authentic as a photomosaic, not just a mosaic
-faster render time

Hi-Res pros:
-more detail on the big pic
-possibility to render more images clearly, not just "tight and light"

Personally, I'm leaning toward keeping the resolution as low as possible because it emphasizes the small images and makes more sense with a small image collection. (Why have 6,400 slots for 3,000 images?) Let me know your opinion. This is your project as much as it is mine! (Somewhere in-between is a valid choice too.) :)
 
Wow, 20megs. I hadn't realised it would be that big...

Well, as you say, there's considerably more detail on the higher res picture, but it does take more time to view the component images (zooming and scanning the final image takes longer). Now that I've had a chance to look at the two side-by-side, personally I feel keeping it at the 40x40 is the better move. It's more true to the format of a photomosaic, as you say, but there's still plenty of detail (at least to my eyes).
 
Back