 |
|
|
|
|
|
 |
| ||||||||
 |
 
|
|
|
|
|
|
|
|
|
| ||||||||
|
|
| 
| User list | You are not logged in.
Pages: 1
It occurs that detection quality for 16bit TIF-s produced by Lightroom 3 x64 and Camera Raw 5.6 x64 is very different. 1.72 vs 3.61
the same XMP files were used for raw processing.
JPG files produced by the same CameraRaw from the same raw files get detected at 1.91 as well.
16bit TIF files produced by RAWTherapee are also detected with high quality 1.86
This may not be APG bug, however, it would be great to have APG detect TIF-s from CameraRaw with higher quality.
When APG is detecting pano from raw files directly, quality is about 1.81
Specific issue with detecting Camera Raw tifs: it seems that images are more distorted and this causes misalignment.
If an issue is be identified with Camera RAW TIFF files, such as some specific tag is missing, etc., this would allow to forward the request to the CameraRaw team.
Camera used: Nikon D700, Lens Nikon AFS 28-70F2.8
Image set had 2 files at 28mm and 1 file at 31mm
Last edited by Michael Ezra (2010-02-19 15:26:26)
Offline
1. RMS is not a great measure of quality of the stitch, as you say "images are more distorted...". BTW, what do you mean by "Camera RAW TIFF files"?
2. noise can fool SIFT, and the different RAW converters may have different noise reduction algorithms. But the more you reduce noise, the less crisp the features are.
Offline
hankkarl wrote:
1. RMS is not a great measure of quality of the stitch, as you say "images are more distorted...".
Geometric distortion of individual images in the panos is very different:
Pano with TIFs from Camera Raw look more bulging and this does not allow them to be aligned properly in the pano.
If, however, pano is made from JPG images produced by the same camera raw from the same raw files (that were sources of TIFs) then individual images align really well and RMS is low.
EXIF in TIF-s and in JPG-s shows identical focal mm. APG also sees correct focal distance in both TIFs and JPG (when looking at image properties)
hankkarl wrote:
BTW, what do you mean by "Camera RAW TIFF files"?
Camera RAW TIFF files = TIFF files created by Camera RAW
hankkarl wrote:
2. noise can fool SIFT, and the different RAW converters may have different noise reduction algorithms. But the more you reduce noise, the less crisp the features are.
Test was run on noise-free images. Besides, I am pointing that there is a Camera RAW-specific issue, as when using the same source RAW files, JPG produced by camera raw produce better pano than TIFs produced by camera raw.
Last edited by Michael Ezra (2010-02-19 17:40:46)
Offline
This appears to be even more interesting.
I opened the problematic TIF files in photoshop and saved as JPG file.
Tried to detect a pano from JPG-s. RMS = 1.91, while RMS from TIFs=3.61
Something is really going on here...
Alexandre, could you take a look?
Last edited by Michael Ezra (2010-02-20 01:42:37)
Offline
JPG is a lossy compression. what did you set the JPG quality to?
All images have noise. Its more visible in the blacks because (we hope) noise is equal to only a few photons--that is, if you have 12 bits (0-4095 discreet levels) noise may add some random number (say, up to 10) to each pixel. There's not much difference between 4000 and 4010, but there is a big difference between 10 and 20. Add to this the fact that our eyes perceive shades exponentially--17% gray is perceived as halfway between black and white (or is it 12%? there's some disagreement). So mid-gray is 696 out of 4096 and 25% gray is 118.
When you say there is no noise, I assume you mean that there is no perceived noise because you shot at ISO 100 (or 200 on a Nikon), exposed to the right, had good lighting and otherwise did the right things. There's still noise, but you don't perceive it. If you increase the EV enough, you will see noise in the blacks.
What JPG does is to throw away some of the information, sometimes it tosses fine differences in color--its not a smoothing filter or a noise reduction filter, but some parts of JPG do smooth (or blur or posterize) the image.
It may be that lightroom does more noise reduction than ACR, but IIRC, you can set the level of noise reduction in both programs.
I don't know if APP prefers brighter areas when running SIFT--if the issue you describe is caused by noise, it may be a good thing for it to do.
BTW, what happens if you run Noise Ninja or similar on the ACR before you output it as TIF?
Last edited by hankkarl (2010-02-21 00:54:07)
Offline
Speaking of image differences, don't forget about the completely different demosaicing methods that can possibly be used by Lightroom and Camera Raw. Because of them, the same raw photo developed using differing demosaicing algorithms can be completely different at the pixel level. Saving an image to a lossy format again is likely to completely change it at the pixel level, but I doubt such small differences would influence SIFT though, someone from Kolor should confirm here, but from my understanding of SIFT it uses larger areas to compare similar features.
It sounds like you have a case here, Michael. It would help the Kolor folk if you could upload your various TIFF and JPG and raw files, either to their ftp, or to a file hosting site.
Offline
There is something else I just discovered.
I developed RAW files into TIF-s using ACR but different ACR settings (exposure, contrast, etc).
Screenshots below illustrate the issue with "curved images in planar projection".
I think this is the reason for bad detection.
Offline
when detection quality is changed to maximum, curved frames in planar projection become straight.
This is very strange - why should detection quality have an effect on this geometrical issue?
Offline
Pages: 1
