Creator of the image here. First of all, thank you for all the attention, it's a real honor to be on the front page oF HN for almost a whole day now.
I am the founder of 360cities.net and creator of previous world record images such as "Prague 18 gigapixels", "Strahov Library 40 gigapixels" and "London 80 Gigapixels". This newest image was commissioned by BT as a kind of closing "Thank you London" gesture of the Olympic year. Of course, they didn't want "just" a jaw-droppingly enormous image, they wanted a world record image! Unfortunately, these things are getting a lot harder to accomplish, and we're starting to go up against some constrainsts of optics, air density, and so on.
So, to create this image, we needed 4 cameras shooting simultaneously in order to get all the images in a reasonable length of time. We needed the very best optics out there, and CMOS sensors with the smallest pixels (somewhat unfortunately this meant the 3 year old Canon 7D). We needed the fasted and most precise panoramic robots to mount the cameras on. We needed gigantic memory cards, 128GB, to store a full run of images on one card. We needed 4 laptops to copy the cards quick enough to shoot again (buying more cards wasn't in the budget). In short, we needed a ridiculous amount of gear - this took 2 days to set up.
We spent 3 days shooting. The first day it was raining, but we shot anyway to learn the gear and see what mistakes we might make. The second day there were 50km/h winds which, 150 meters up, is very cold and a little bit dangerous. We made sure the gear was triple secured, and we shot anyway. One of the three photographers threatened to leave the project. It was unpleasant. But we uncovered more issues which we were able to solve. Keep in mind the cameras were moving continuously, and focusing, and shooting - they never stopped to take a shot. 12600 shots while moving and focusing each time? This required some very exacting work, knowing the gear, and maybe even a bit of luck.
Day 3 was clear, and not windy. We got 2 sets of images shot. All cameras got the shots they needed in 2 sets. That is 96000 images with not a shot missed. This to me is a huge technological achievement that i'm proud of, and I will sing the praises of Canon anytime someone asks me to :-)
Now, on to stitching. Normally I have no problem stitching 2 or 4 thousand images. But this is an order of magnitude more. As gabriel weinberg noted recently, you find out what breaks when you increase things by an order of magnitude.
Basically, the stitching software, Kolor Autopano Giga, was basically inoperable with this size data set. They scolded me for not having told them about this shoot beforehand. They told me that they know about the issues but other things are taking precedence in the development of the software. Stitching a world record amount of images is, by definition, an edge case. So they weren't really able to help me.
Stitching ended up taking nearly 4 months when we thought it would take about 4 weeks. These 4 months were spent working until 3 or 4am many nights of the week. Deadlines slipped. I suffered. My family suffered. After 6 weeks of work I had accomplished nearly nothing and started to panic. I just had to keep on pushing....
In the end, we had to set a delivery date for the client. This required me to leave a few mistakes in the image that I would have liked to fix. However keep in mind that some areas of bad alignment on a normal image might take 30 minutes to fix, but with this size project in the software, it took A WEEK to fix. This kind of work is extremely demoralizing and there isn't much to be learned from it either.
In the end, the image is finished. As they say "you never finish, you only stop!". I'll probably make a few more fixes, later. For right now I need to decompress for a while.
So that is some background on how the image was created. It's not perfect, but overall it is much better than I expected. Shipping it was a terrible, brutal grind that I hope to never experience again. But it's done now and that feels pretty sweet.
I'm typing this on the bus home from the airport after being in London for a tv show where they announced the image. That was fun. But the best feeling is knowing that the thing is finished and shipped, and if I ever have to do a similar job again I will hopefully learn from the mistakes made on this one and not have such a terrible grind again.
I'm happy to answer any questions about the image that I didn't answer here. thank you!
* The image has too much unsharp mask applied, which unfortunately is quite noticeable. The buildings look too sharp, too much edge contrast from over-sharpening.
* The HDR effect looks too obvious and forced. This image is definitely not a good candidate for HDR. It simply wasn't needed, and reduces the quality of the image.
* You've inverted the mouse scrolling, it's the wrong way around! Clicking and dragging with the mouse to the left of screen should move the image to the right, as it does on Google street view - this is the way panoramas are mostly done. There's no right or wrong, but it's confusing to have the opposite of what is conventional.
OK, I have heard quite a lot that I used too much USM (unsharp mask). Fine, I will tone it down on the next iteration. Without any USM, the image just looked a bit foggy, washed out, and not very nice at all. I went a bit overboard on it, I agree with you. It is very difficult to get a balance with an image of this resolution, because when you're zoomed all the way in, I think the amount of USM is ok (I like the stark contrast with a lot of blacks and a lot of whites) but when you zoom out, it looks a bit yucky. With a lower resolution image there isn't as much difference in tonal contrast between the most zoomed in and zoomed out tones.
There is no HDR effect at all - it is only the USM artifacts that I speak of above.
Mouse - see my other comment below - we are using the "original" QTVR (and video games) style scrolling which is more "Virtual Reality" than the "click and drag" that has now become standard. Again, I agree, we will start using the "new standard".
Amazing work. It's incredible (and a bit scary in a big brother way) to be able to zoom-in CSI-style across such a large distance and still see faces / license plates, especially with mostly consumer/pro-level equipment on an ad-budget.
Very interesting! Sounds like a monumental effort! Though if it was only the stitching that took the vast majority of time and effort, no doubt you or others should be able to break this record yet again in time as software progresses?
How long (and when - start date and time) did it take to get all those 96,000 images on Day 3, please?
Since I've had nothing better to do that look at the pictures for a bit, I'm confident that it was Sunday, 28th October 2012, with one set of photos starting around 1pm facing north and another one used for filling in gaps a bit later - starting somewhere between 3pm and 4pm (harder to be precise - there are fewer clocks visible from this one).
Excellent work! You can feel comfortable that you will have a long lasting legacy; it's a real snapshot of history, imagine having one of these from 1912. Would it be possible to see one or two of the individual images (to get a better perspective)?
Thank you very much. This is actually a comforting thought. I hope the image actually does stick around far enough into the future for it to have the value you suggest. So much of what we create in the world of computers and software seems more like performance art than anything truly long-lasting.
I'm incredibly impressed. I was able to see the brand of someone's laundry detergent through a window. I've never seen that type of zoom before one one of these.
1) we used a canon 400mm f/2.8 L ii lens with a canon 2x teleconverter (mark 3)
2) the images get more "foggy" at the max zoom level because at the smaller magnification, I was able to apply some filters to the image (unsharp mask, mostly) to improve the local contrast. At the deepest zoom level, this is not possible, because the source files at this level were too large to open in Photoshop. I have applied some color corrections to the source images, and I am re-rendering these files, but I didn't make the deadline to put this live. It will go live in a few weeks probably.
I noticed the same thing. I think the max zoom level are the images have original color settings and mostly untouched while the non-max zoom pictures have been color calibrated (more contrast and such).
did you guys try to stitch in batches, iteratively? like a few k at a time, then stitching the larger ones together? it seems like there is a lot of room for improvement if you can feed the photos in a pre-sequenced order also, rather than just throwing the full batch in and treating them like a giant jigsaw.
Well, the image needs to be optimized all at once in order to have the correct shape. Especially since it is a sphere. You can't really cheat so much.
Re: importing in a pre-sequenced batch: YES this works on a few hundred or thousand images. But it didn't work with these images. The software broke. It didn't work. :-(
Sorry, this panorama uses the original "qtvr" style navigation. Google streetview, maps, and Earth use "click and drag" navigation. There is currently a religious war going on in the pano community about which one is better. I tend to agree that the "click and drag" has become the standard.
I guess I don't get what's so horrible about giving people a choice. This is just a Quake-style mlook implementation at the end of the day, and gamers are used to being able to select the mlook polarity.
But no, there has to be a jihad, and somebody has to "win."
Of course! I walk past that road everyday so could make an educated guess at 40km/hr. Mind you, my maths skills are poor so I won't be about to start delving any deeper into the calculation!
EDIT: I'm wondering - if Google has to anonymise faces and registration plates on street view, shouldn't BT be doing it too with this high-res panoramic image?
Google doesn't HAVE to blur faces (except in only about two countries). They do it as a PR gesture, to please people. This has had terrible effects for photographers, who are often told that "they have to blur faces like Google does".
Ah, you beat me too it. That's exactly the link I was trying to paste. I thought the URL updated automatically, I didn't realise you needed to click share.
Readit News
Creator of the image here. First of all, thank you for all the attention, it's a real honor to be on the front page oF HN for almost a whole day now.
I am the founder of 360cities.net and creator of previous world record images such as "Prague 18 gigapixels", "Strahov Library 40 gigapixels" and "London 80 Gigapixels". This newest image was commissioned by BT as a kind of closing "Thank you London" gesture of the Olympic year. Of course, they didn't want "just" a jaw-droppingly enormous image, they wanted a world record image! Unfortunately, these things are getting a lot harder to accomplish, and we're starting to go up against some constrainsts of optics, air density, and so on.
So, to create this image, we needed 4 cameras shooting simultaneously in order to get all the images in a reasonable length of time. We needed the very best optics out there, and CMOS sensors with the smallest pixels (somewhat unfortunately this meant the 3 year old Canon 7D). We needed the fasted and most precise panoramic robots to mount the cameras on. We needed gigantic memory cards, 128GB, to store a full run of images on one card. We needed 4 laptops to copy the cards quick enough to shoot again (buying more cards wasn't in the budget). In short, we needed a ridiculous amount of gear - this took 2 days to set up.
We spent 3 days shooting. The first day it was raining, but we shot anyway to learn the gear and see what mistakes we might make. The second day there were 50km/h winds which, 150 meters up, is very cold and a little bit dangerous. We made sure the gear was triple secured, and we shot anyway. One of the three photographers threatened to leave the project. It was unpleasant. But we uncovered more issues which we were able to solve. Keep in mind the cameras were moving continuously, and focusing, and shooting - they never stopped to take a shot. 12600 shots while moving and focusing each time? This required some very exacting work, knowing the gear, and maybe even a bit of luck.
Day 3 was clear, and not windy. We got 2 sets of images shot. All cameras got the shots they needed in 2 sets. That is 96000 images with not a shot missed. This to me is a huge technological achievement that i'm proud of, and I will sing the praises of Canon anytime someone asks me to :-)
Now, on to stitching. Normally I have no problem stitching 2 or 4 thousand images. But this is an order of magnitude more. As gabriel weinberg noted recently, you find out what breaks when you increase things by an order of magnitude.
Basically, the stitching software, Kolor Autopano Giga, was basically inoperable with this size data set. They scolded me for not having told them about this shoot beforehand. They told me that they know about the issues but other things are taking precedence in the development of the software. Stitching a world record amount of images is, by definition, an edge case. So they weren't really able to help me.
Stitching ended up taking nearly 4 months when we thought it would take about 4 weeks. These 4 months were spent working until 3 or 4am many nights of the week. Deadlines slipped. I suffered. My family suffered. After 6 weeks of work I had accomplished nearly nothing and started to panic. I just had to keep on pushing....
In the end, we had to set a delivery date for the client. This required me to leave a few mistakes in the image that I would have liked to fix. However keep in mind that some areas of bad alignment on a normal image might take 30 minutes to fix, but with this size project in the software, it took A WEEK to fix. This kind of work is extremely demoralizing and there isn't much to be learned from it either.
In the end, the image is finished. As they say "you never finish, you only stop!". I'll probably make a few more fixes, later. For right now I need to decompress for a while.
So that is some background on how the image was created. It's not perfect, but overall it is much better than I expected. Shipping it was a terrible, brutal grind that I hope to never experience again. But it's done now and that feels pretty sweet.
I'm typing this on the bus home from the airport after being in London for a tv show where they announced the image. That was fun. But the best feeling is knowing that the thing is finished and shipped, and if I ever have to do a similar job again I will hopefully learn from the mistakes made on this one and not have such a terrible grind again.
I'm happy to answer any questions about the image that I didn't answer here. thank you!
The main problems are:
* The image has too much unsharp mask applied, which unfortunately is quite noticeable. The buildings look too sharp, too much edge contrast from over-sharpening.
* The HDR effect looks too obvious and forced. This image is definitely not a good candidate for HDR. It simply wasn't needed, and reduces the quality of the image.
* You've inverted the mouse scrolling, it's the wrong way around! Clicking and dragging with the mouse to the left of screen should move the image to the right, as it does on Google street view - this is the way panoramas are mostly done. There's no right or wrong, but it's confusing to have the opposite of what is conventional.
OK, I have heard quite a lot that I used too much USM (unsharp mask). Fine, I will tone it down on the next iteration. Without any USM, the image just looked a bit foggy, washed out, and not very nice at all. I went a bit overboard on it, I agree with you. It is very difficult to get a balance with an image of this resolution, because when you're zoomed all the way in, I think the amount of USM is ok (I like the stark contrast with a lot of blacks and a lot of whites) but when you zoom out, it looks a bit yucky. With a lower resolution image there isn't as much difference in tonal contrast between the most zoomed in and zoomed out tones.
There is no HDR effect at all - it is only the USM artifacts that I speak of above.
Mouse - see my other comment below - we are using the "original" QTVR (and video games) style scrolling which is more "Virtual Reality" than the "click and drag" that has now become standard. Again, I agree, we will start using the "new standard".
How long (and when - start date and time) did it take to get all those 96,000 images on Day 3, please?
Note: USB 3 card readers (that we were able to buy) are not really "usb 3" :-(
Why do the images get more "foggy" at the max zoom level?
Do you have any photos of the robots operating the cameras?
1) we used a canon 400mm f/2.8 L ii lens with a canon 2x teleconverter (mark 3)
2) the images get more "foggy" at the max zoom level because at the smaller magnification, I was able to apply some filters to the image (unsharp mask, mostly) to improve the local contrast. At the deepest zoom level, this is not possible, because the source files at this level were too large to open in Photoshop. I have applied some color corrections to the source images, and I am re-rendering these files, but I didn't make the deadline to put this live. It will go live in a few weeks probably.
(Disclaimer: I'm not a photographer)
Re: importing in a pre-sequenced batch: YES this works on a few hundred or thousand images. But it didn't work with these images. The software broke. It didn't work. :-(
1. The Shard (Tate Modern in the foreground) - http://btlondon2012.co.uk/pano.html?view.hlookat=-33.1601...
2. London Eye - http://btlondon2012.co.uk/pano.html?view.hlookat=-3.8135&...
3. Tower Bridge - http://btlondon2012.co.uk/pano.html?view.hlookat=-38.7356...
4. Wembley Stadium - http://btlondon2012.co.uk/pano.html?view.hlookat=140.7253...
5. Battersea Power Station - http://btlondon2012.co.uk/pano.html?view.hlookat=35.2192&...
6. Olympic Park - http://btlondon2012.co.uk/pano.html?view.hlookat=-73.9268...
7. Big Ben & Houses of Parliament - http://btlondon2012.co.uk/pano.html?view.hlookat=6.4768&...
8. MI6 Building - http://btlondon2012.co.uk/pano.html?view.hlookat=14.9167&...
9. St Pauls Cathedral - http://btlondon2012.co.uk/pano.html?view.hlookat=-43.8561...
10. Nelsons Column - http://btlondon2012.co.uk/pano.html?view.hlookat=13.4841&...
11. Behind the iconic billboards in Picadilly Circus - http://btlondon2012.co.uk/pano.html?view.hlookat=16.9026&...
12. Harrods Department Store - http://btlondon2012.co.uk/pano.html?view.hlookat=64.6214&...
13. The British Museum - http://btlondon2012.co.uk/pano.html?view.hlookat=-45.7275...
And Wimbledon tennis (the AELTC, 7 miles) - http://btlondon2012.co.uk/pano.html?view.hlookat=58.1529&...
http://btlondon2012.co.uk/pano.html?view.hlookat=-48.0944...
But no, there has to be a jihad, and somebody has to "win."
Interesting to see a shot capturing the progress of a car which helps determine the rate at which the images were taken.
http://btlondon2012.co.uk/pano.html?view.hlookat=168.9705...
http://btlondon2012.co.uk/pano.html?view.hlookat=174.1983...
EDIT: I'm wondering - if Google has to anonymise faces and registration plates on street view, shouldn't BT be doing it too with this high-res panoramic image?
I don't remember architecture like this! :P
edit: Wrong link. -_- Intended to post upside-down building...
edit2: Corrected link.
Also, this guy?