Thought I'd run the images through 2.6.0b4, and I have to say that the improvements in stitching are huge. APG has always struggled with this set previously due to the repeating patterns of the floors (to the extent that it was simply too much like hard work to fix it manually), but I only had to lend a hand to correct 4 image pairs for this.
Thanks for the feedback, gddxb. I happy to see, we are going to the right direction.
Yes, we are finishing to patch the blending in the sky. Probably, the most complicated issue to solve. One tip : doing a rendering this way : uncheck anti-ghost, check diamond weighting, multiband, does work well on sky, but not well on ground. Do a second rendering with default setting for ground, and blend manually on photoshop to get a perfect result. A bit complicated, I agree. We want the same result to be achievable with default settings.
@bo - I may be wrong, but my perception is that blending of the sky has improved with recent releases too, perhaps Alexandre would like to comment?
One thing is for certain though, speaking as someone who has bought, and regularly tries out, all the major panorama stitching programs out there, Autopano remains the best solution for the kind of panos that I shoot. There are aspects of the other solutions that I would like to see in Autopano - particularly with regards to the layout of input photos (awaiting a firmware update from my pano head manufacturer that will probably negate the need for Kolor to do anything on that front though), but in my experience, when it comes to the production of the output panorama, Kolor are in a league of their own.
Great to see that the blending issue is recognized, and that it is being addressed.
APG 2.60 beta 4 There is more work to be done. The example below show trouble in the sky, still. It is shot with a Olympus 50mm stopped down to f8.0 on my E-P2 (100mm equiv) . A sphere of 376 images. Rendered this way : "uncheck anti-ghost, check diamond weighting, multiband" I don't think its vignetting. btw: I have unchecked all color-altering options.
Quite for sure, this is vignetting. It gives these typical darken zone at corners of images.
Hi Alexandre -
It may not be. I recently shot a 360 spherical of the Sheikh Zayed Grand Mosque (see other thread) and you'd think looking at the sky that there was vignetting. However, I was shooting with a Canon 90mm TS-E on a 1D Mk IV. The vignetting on this lens on a crop camera at the aperture I was shooting is practically non-existent:
Ok. I want to check what vignetting looks like. So I did this experiment: - I used an image from the shoot, erased it, filled it with gray color in Photoshop and imposed an extreme vignetting - the gray vignetted "image" is shown below as image 3 - then I made 376 copies with filename identical with the images from the above example - I swapped the originals with the new gray vignetted "images" - APG accepted the decoys and I rendered it - image 1 below shows the "detected" preview - image 2 below shows the rendered result
Does this result conclude that AlexandreJ "Quite for sure, this is vignetting." is right ? I'm not sure. What about the vertical bands, that are missing ?
I will do an additional exercise before concluding: de-vignette the original images in DXO and rerun the rendering.
I have done some more experimenting: - I have de-vignetted the images i DXO, in strength 20 and 50 - the result on an image from the sky shown below - then I swapped in the de-vignetted images and rendered the pano - examples from the rendered pano shown below
The verdict: there are dark bands in the sky even with the extremely "de-vignetted 50" what's the problem? I don't know is it the lens alone? I don't think so
Kolor should take another look at this. As more motors get available there will be more people shooting giga-spheres, which will include the sky. Then there will be more skies with dark bands and more posts about the phenomena.
Okay. Not easy to conclude. I did the following test to measure a little bit more the vignetting part in the provided image. Vignetting is symmetric, so when rotating the image, it should be reflected the same in any direction. So I took the provided image and made symmetric version of the same image and aligned them around the original image. If the image was suffering only from vignetting, you would see no seam at edge ( you can do the test with the grey vignetting image generated ).
So, it shows that images have a color gradiant with a direction, but also some vignetting but not strong it seems. I don't know what to think yet and to make conclusion, it would require for example a small extract of the panorama in the sky like 4 images forming a 2x2 matrix. This could help to measure the gradiant component compared to the vignetting component. Could you provide us 4 full resolution images from the panorama ( from the sky of course ) ?
BTW : I increased the rendering to make the default stronger. Removing the vignetting does help a little, but as the vignetting function is not well calculated, I assume it just removes the first level of issue.
One question : which autopano version and what rendering settings ?
There is a package of TIFF files for download at www.rundskuer.no/ymse/kolor.zip (the images were shot as ORF and converted to TIFF using Olympus Master software) It contains 8 images from the sky as shown below. The images close to zenith is stretched a lot in the resulting pano, also shown below. To pinpoint where in the sphere the images were shot I have included the XML used for import.
I hope this can be a help to solve the dark-band-in-the-sky problem. I will also shoot some spheres with other lenses to see if there are more or less bands.
I used your package to create a .pano project file that just contains this extract. I did 2 renderings in PSD format with embedded images to look at the vignetting. On the top of each layer, I added a level layer to raise drastically contrast on analysed zone. Here's the results : * first screenshot : just some images revealed. When you look at corners, obviously, there's vignetting, but it rapidly drops after 10% inside the image. The formula of this vignetting here has a quite quick reduction. * second screenshot : that's the rendering done, with every settings by default on the 2.4 RC1 engine. It seems that the fixes made reduces a little the issue we can see in your rendering and not in this one. * third screenshot : no anti-ghost, diamond weighting. This has always been the right solution for nice skies, but it isn't good at all for ghost or ground. You can guess it by looking at tree top.
Perhaps, we can go more deeply in the vignetting measurement. Adobe released a tool whose aim is about that : adobe lens profile creator. It does measure lens distortion but also vignetting effect. If you remember the lens and aperture used, we could try to make some shots in the same condition on a white paper to measure the vignetting well. Then use this profile in either photoshop to postprocess the images or in autopano through the open lens plugin. I think it will solve some of the issues, but not all of them. There's a gradiant in these images I cannot explain and I don't think it comes only from the sky natural gradiant.
So Glad to see this thread! Depending on the length of capture the color can shift too. Late morning to afternoon blue can change hue and intensity. So, there are other elements to consider in the gradation. Could the solution be to have a color anchor that states this a a gradation color? I'm sure similar issues would be found in other similar color situations, a white room for example. I usually recreate the skies in PS. Visually the sky should look homogeneous and infinite. If there was a why to tell the stitcher to look at the entire sky and equalize it. When the render comes along in its multi band path, it can recall the newly calculated equalization of the entire sky.
I appreciate very much the interest in users troubles from Kolor ! That makes me want to give something back, to make APG better.
And I'm still struggling with the sky. This time I shot a sphere using Olympus E-P3 with a Panasonic Leica DG Summilux 25mm f1.4 ASPH lens. This is a prime lens and tests show very little vignetting stopped down to f2.8. I shot the sphere at f8.0.
In my search for a smooth sky I tried three different rendering settings: - simple - multiband diamond - antighost
The "simple" setting gives me the smoothest sky, but even there I can see patterns from the single images Multiband shows more patterns and antighost shows even more. The "simple" setting cannot be used if there are anything moving in the pano, without a lot of postprocessing.
Is there anybody who has been tweeking the the blending setting and found a good combo for a smooth sky ?
There's another parameter you can play with : it's the depth of the multiband : the deepest, the longest the zone of blending. By default, in beta 4, it's -2. You can use 0 instead to raise that level by 2, which is in fact 4x more long. With that tip, the multiband diamond should be the best, followed by antighost.
As we made some small adjustement in this code, you can also wait for RC1 ( in 2 days from now ), which shows better results ( my screenshot don't have as much issue as you have ).