Digital Photo Professional - New release 4.21.30

[March411]

For those that use it it's available here: Digital Photo Professional 

When you use the link it should recognize your OS and supply the correct download. There isn't much in the way of release notes but they can be found here: Digital Photo Professional 4.21.30 for Windows 

I have loaded it and taken it for a spin. Noting much to note although it does appear to load images faster. I always had an issue where the photo would load and then a delay to focus, that time has been reduced.. 

[JamesHarvey]

Yes, if you want to see optimum resource use you should look at your PCs performance manager when running DaVinci Studio.  I often look where I am in rendering, knowing what effects etc. I have applied to each clip & correlating that with the CPU & GPU components' activity.  I hope the next major update to DPP has simalr optimized dynamic resource use. /JH

[ebiggs1]

"I remember Hallmark in Kansas City."

Hallmark is still in KC just not any resemblance of its former self. In fact KC intends on building the new KC Royals baseball stadium where the main office facility currently is. 

"What I like about DPP is the "digital lens optimizer" and the DPRAW tool. I do not know of other programs that do those things as well."

However, that is an opinion and I, too, think DPP4 is probably the most accurate but I still prefer Adobes as do almost every pro shop in the world like Hallmark Cards of Kansas City does.

[johnrmoyer]

The biggest weakness to Adobe products is that so far as I know, they do nothing about small aperture diffraction blur. Adobe only corrects vignetting, distortion, and chromatic aberration so far as I know. Correcting diffraction blur requires a lot of computation time is likely the reason.

If one always uses a medium format sensor or always uses a sensor with low pixel count or always uses a wide aperture then the diffraction blur will not be a problem. 

[ebiggs1]

"...small aperture diffraction blur."

Well, for sure, that isn't a setting or drop down and I also don't remember if DPP4 has such a drop down either. However, this is easily corrected in PS. Perhaps not as completely as whatever the control is in DPP4 but I already acknowledged the superiority of DPP4 for lens correction. I mean it should it is Canon's own stands to reason. Right?

This is a personal preference but myself and like I said virtually 100% of all professional shops prefer and use Photoshop. If you convert the layer to a smart object (which you should always do), Filter, Lens Correction make manual corrections and record it as an Action. It will be a single click from then on.

[johnrmoyer]

---

@ebiggs1 wrote:

"...small aperture diffraction blur."

Well, for sure, that isn't a setting or drop down and I also don't remember if DPP4 has such a drop down either. However, this is easily corrected in PS. Perhaps not as completely as whatever the control is in DPP4 but I already acknowledged the superiority of DPP4 for lens correction. I mean it should it is Canon's own stands to reason. Right?

This is a personal preference but myself and like I said virtually 100% of all professional shops prefer and use Photoshop. If you convert the layer to a smart object (which you should always do), Filter, Lens Correction make manual corrections and record it as an Action. It will be a single click from then on.

---

In DPP diffraction correction is in the same menu area as digital lens optimizer. Since it is included in digital lens optimizer, it is grayed out when digital lens optimizer is enabled. In Rawtherapee, correction of small aperture diffraction blur is done in "capture sharpening". Rawtherapee will do it for any lens. Canon DPP will only do it for Canon lenses and maybe only cameras that have it built into the camera. When using a 3rd party lens, Adobe software will have an advantage over DPP. DPP will use knowledge of the sensor chip in the camera and knowledge of the lens to calculate the radius parameter for diffraction blur correction. With Rawtherapee, one must do the calculation oneself. With the GMIC commandline program, one must also calculate the radius onself. ( I made myself a table listing the radius for each F Number of my EOS 80D and later for my EOS R5 ). The Richardson/Lucy deconvolution was originally invented for telescopes and was later also used for microscopes and cameras.

Small aperture diffraction blur correction is important for cameras with a small sensor area or with photo sites close together on the sensor chip. For EOS R5 diffraction blur starts at about F/7.1.

[ebiggs1]

Way above my head. I'm just a simple photographer. I've stated my case.

[JamesHarvey]

Have you compared correction for diffraction in DPP with that in Canon Neural Network?  I have found it to be very successful on astro (mainly lunar) images taken with the R7 & RF200-800.  It's also interesting that NN will not apply lens correction to images taken with the RF800/11 or 600/11, though it will work with other consumer teles, maybe because those lenses use DO, whereas DPP appears to have no such overt restriction?  Thanks! 

[johnrmoyer]

---

@JamesHarvey wrote:

Have you compared correction for diffraction in DPP with that in Canon Neural Network?  I have found it to be very successful on astro (mainly lunar) images taken with the R7 & RF200-800.  It's also interesting that NN will not apply lens correction to images taken with the RF800/11 or 600/11, though it will work with other consumer teles, maybe because those lenses use DO, whereas DPP appears to have no such overt restriction?  Thanks! 

---

I have not tried the Canon neural network. I have played with the free neural network in GMIC.

I am just guessing. Richardson/Lucy requires aperture shape, FNumber, spacing of photosites on the sensor chip, and wavelength of light for the calculation. Rawtherapee where I can see the source code assumes that the aperture shape may be approximated by a circle and that the central peak of the Airy disk may be approximated by a Gaussian. The wavelength might be approximated by using the wavelength for green light, but that will introduce some error at each iteration when there is a mixture of wavelengths.

To the extent those assumptions are met, the Richardson/Lucy does not add error at each iteration but converges to an exact correction of diffraction blur. The assumptions are never met perfectly because the aperture is not in fact perfectly round and the Airy disk central peak is not quite Gaussian and there is usually a mixture of wavelengths. If the image were mostly red instead of green then using the green wave length to calculate the Airy disk diameter would give the wrong diameter. To the extent there are various colors in the image, there will be a different diameter for each wavelength. Using luminance for the Richardson/Lucy deconvolution makes different assumptions about the mixture of light wavelengths and sometimes gives a better result, but then the question becomes which way to calculate luminance. I have played with treating the raw camera data as if it were photon counts and as if it approximates luminance and that might work for astro. 

Noise in the image also violates the wavelength assumption for Richardson/Lucy. The Poisson distribution of photon counts and energies does not map neatly to wavelength. For a small number of photons there is greater probability of an unexpected color. DPP and the Canon cameras reduce the amount of "digital lens optimizer" when there is higher ISO. My guess is that this reduces the number of iterations so that the error is less noticeable.  

I might guess that Canon has mathematical models of lens design that can be used to calculate a more accurate point spread function of the Airy disk for each lens and can use lens design information to improve a Richardson/Lucy deconvolution in "digital lens optimizer". This would be more accurate than approximating it with a circle and a Gaussian. 

In contrast, instead of calculating the behavior of photons, the neural network was trained on a large number of images. I might guess that if there were not enough training images then the neural network would not be available.

 https://en.wikipedia.org/wiki/Photon_statistics

[ebiggs1]

One thing I will add being a very long time PS user, anything that DPP4 can do editing a photo can be done in PS. It may not be a drop down or a single click granted.  But a lot of what can be done in PS is not even remotely possible in DPP4. That may change in future editions of DPP4 but not today.

[JamesHarvey]

Thanks for the detailed explanation.  I have a professional background in signal processing & physics which enables me to understand a lot of what look like black boxes to most photographers.  I had not looked at the Richardson-Lucy deconvolution much before but in reading it again I'm beginning to see how that would produce different results to statistically-trained processes, and how those differences would be influenced by optical design.  Thanks for the pointer to RawTherapee; definitely worth a look for me given the options & separation of processes that it provides.