DPP3 far outperforms DPP4 (an ISO invariance experiment)

[Stemanz]

Everyone seems pretty much happy with DPP4.

I'm not.

A simple ISO-invariance experiment, performed with an EOS 1200D, shows tremendous detail loss when the same .CR2 images are processed with DPP4 compared with DPP3.

My procedure is simple: shooting a picture (tripod, indoor) with varying exposure times and ISOs (aptly chosen), then boosting the EV in post-processing to match the picture taken on camera. Also, the very same noise reduction profile has been applied to all same-ISO-equivalent images (that is, the exact noise reduction profile that DPP* would apply to an in-camera ISO 3200 picture is applied in post-production to a +1EV ISO 1600 picture, and so on).

DPP4 processing is lagging behind at +1EV already. Things are just going to be getting worse:

Clearly, DPP4 is doing a horrible job in processing the boosted CR2 (top right) while DPP3 renders almost in par with the natively taken ISO 6400 image the +2EV boosted ISO 1600 CR2 file (top left).

What is your experience? Is anyone having these issues? How is DPP4 doing with better cameras than mine?

If anyone asks, I will be more than happy to provide full size CR2 files to experiment with (the full image has darker and brighter areas, I just wanted to keep it simple in here).

And NO, I'm not going to use Lightroom/RAWtherapee, etc. I want DPP4 to do a decent job out of the box.

[Waddizzle]

If you used a tripod, then why are your ISO set so high?  The EOS 1200D does not produce many usable photos once get above ISO 800.  What is the point of adding +1 Ev in post, anyway?

[EDIT] A more meaningful test would have been to use 3 stops of exposure correction.  Using Manual shooting mode, take a good exposure at ISO 800.  Then dial the ISO down to 100, and take another shot.  

Now use DPP to add +3 Ev to the ISO 100 exposure, and compare which one did a a better job of matching the ISO 800 exposure.

[ebiggs1]

"I want DPP4 to do a decent job out of the box."

Good luck!

"I'm not going to use Lightroom/RAWtherapee,..."

Too bad!

[shadowsports]

stemanz,

We see what you've done, but you have to admit this is a very obscure and extreme usecase. 

Clearly you are aware that pushing the Digic 4's ISO to these levels is going to cause some issues which will present in any post processing software. 

I think Wadizzles points should be considered.

Since DPP is enthusiast grade at best, getting a pro (ebiggs1) to bite on its overall usefulness considering what else is available isn't likely to happen.

Admittedly, I doubt I would ever use a camera in these types of extreme shooting situations.  My understanding of exposure and knowing my hardware's DR limitations would kick in long before I would try capturing images under these conditions.   

Sometimes older equipment does run better with older software.  Eventually all manufactures shift their focus and optimze their current offering to run best with newer equipment.  I don't have any problems with DPP4 and my body (newer than yours).        

That said, I don't use DPP in my primary workflow since I own other lenses.  So I am probably not the best person to make a comprehesive evaluation of its DNR capabilities in photos taken at high ISOs.  Others here can likely help though.

[Peter]

---

@Stemanz wrote:

Everyone seems pretty much happy with DPP4.

I'm not.

A simple ISO-invariance experiment, performed with an EOS 1200D, shows tremendous detail loss when the same .CR2 images are processed with DPP4 compared with DPP3.

My procedure is simple: shooting a picture (tripod, indoor) with varying exposure times and ISOs (aptly chosen), then boosting the EV in post-processing to match the picture taken on camera. Also, the very same noise reduction profile has been applied to all same-ISO-equivalent images (that is, the exact noise reduction profile that DPP* would apply to an in-camera ISO 3200 picture is applied in post-production to a +1EV ISO 1600 picture, and so on).

DPP4 processing is lagging behind at +1EV already. Things are just going to be getting worse:

Clearly, DPP4 is doing a horrible job in processing the boosted CR2 (top right) while DPP3 renders almost in par with the natively taken ISO 6400 image the +2EV boosted ISO 1600 CR2 file (top left).

What is your experience? Is anyone having these issues? How is DPP4 doing with better cameras than mine?

If anyone asks, I will be more than happy to provide full size CR2 files to experiment with (the full image has darker and brighter areas, I just wanted to keep it simple in here).

And NO, I'm not going to use Lightroom/RAWtherapee, etc. I want DPP4 to do a decent job out of the box.

---

Please add the raw files or send them to me via a filehost in PM.

[Stemanz]

Thank you all for your feedback.

I am perfectly aware of the limitations of my camera, but this experiment is not strictly related to any real use case: it just shows how - under some conditions - DPP4 suffers from a very bad regression compared to DPP3 (that's what I am complaining about, shadowsports).

To me, it's OK that my camera's ISO 1600 shots pushed to +2EV do look terrible. I just want them to look equally terrible with DPP3 and DPP4, and not much more terrible with the newer, and supposedly better, DPP4.

Besides, it falls exactly in the ballpark of my usual use case (if you're interested: https://www.flickr.com/photos/44447189@N08/).

All in all, that DPP4 does better with newer equipment is promising, even if not completely satisfying. There's some truth in shadowsport's words: "Sometimes older equipment does run better with older software."

Waddizle, I can't push +3EV with DPP3 for a comparison with DPP4, just +2EV. And I'm more interested in what happens at higher ISOs.

Here you can find (for 7 days) the RAW files and the processed JPGs: https://we.tl/t-DVn1APmlWG

Again, thank you for your time and feedback.

[Waddizzle]

@Stemanz - I think your tests neither prove or disprove anything.  You working with garbage in, and getting garbage out.  I suggest that you retry your tests at the ISOs that I suggested.  Use Av mode to come up with a set of exposure values to use in Manual mode.  You should discover that Canon sensors are not ISO invariant.

I have not used DPP3 in years.  I'm surprised to hear that it only allows 2 Ev of exposure adjustment to RAW files.  I just noticed that my laptop no longer has DPP3 installed, so I was not able to check it for myself.  I guess one of the DPP4 updates might removed it.  ???  I am not a DPP user because DPP does not perform lens correction on non-Canon lenses.

BTW, I looked at several dozen of your Flickr photos.  They are all very nice.  But, I did not see one that may have required ISO values above 800, not even the star trail shots. I almost never shot my T5/1200D using an ISO over 800.  

This is a panorama shot with my T5 and a Rokinon 85mm T1.5 lens.  The ISO was 100, I am certain.  I think exposures were around 8 seconds, and aperture was f/8.

I often hear from people who know what they are talking about when it comes to astrophotography that the best S/N ratios from Canon image sensors is obtained at ISO 800.  They recommend shooting the night sky at ISO 800.  If you are not getting enough light, then either a longer exposure or a wider aperture might be needed.

[ebiggs1]

"I just want them to look equally terrible with DPP3 and DPP4, ..."

Huh?    I guess there must be some logic there.

[Peter]

---

@Stemanz wrote:

Thank you all for your feedback.

I am perfectly aware of the limitations of my camera, but this experiment is not strictly related to any real use case: it just shows how - under some conditions - DPP4 suffers from a very bad regression compared to DPP3 (that's what I am complaining about, shadowsports).

To me, it's OK that my camera's ISO 1600 shots pushed to +2EV do look terrible. I just want them to look equally terrible with DPP3 and DPP4, and not much more terrible with the newer, and supposedly better, DPP4.

Besides, it falls exactly in the ballpark of my usual use case (if you're interested: https://www.flickr.com/photos/44447189@N08/).

am perfectly aware of the limitations of my camera, but this experiment is not strictly related to any real use case: it just shows how - under some conditions - DPP4 suffers from a very bad regression compared to DPP3 (that's what I am complaining about, shadowsports).

To me, it's OK that my camera's ISO 1600 shots pushed to +2EV do look terrible. I just want them to look equally terrible with DPP3 and DPP4, and not much more terrible with the newer, and supposedly better, DPP4.

All in all, that DPP4 does better with newer equipment is promising, even if not completely satisfying. There's some truth in shadowsport's words: "Sometimes older equipment does run better with older software."

Waddizle, I can't push +3EV with DPP3 for a comparison with DPP4, just +2EV. And I'm more interested in what happens at higher ISOs.

Here you can find (for 7 days) the RAW files and the processed JPGs: https://we.tl/t-DVn1APmlWG

Again, thank you for your time and feedback.

---

I compared in DPP3, DPP4, RawTherapee and darktable. As you wrote there is a difference between DPP3 and DPP4.

7 in luminance noise reduction and 10 in chrominance noise reduction are not the same. The difference is in luminance noise reduction. Lower the luminance noise reduction in DPP4 and it will match the higher luminance noise reduction in DPP3.

[Stemanz]

• Waddizzle, thank you for taking the time of browsing through my pictures.

You're right, I usually never had to go beyond ISO 800, most of the time it was more than enough. And I got it all wrong last time, where I had to severely reduce the exposure to get a decent trail, and clipped into white the color of the majority of the stars. I still have a lot to learn.. and remember.

Trouble is: 20" at ISO 6400 was almost required to get anything out of this unremarkable spot of the Milky Way (waiting for the end of spring to get the central juicy part right). And I still have the slow 18-55 f/3.5-6.5 that came with the camera body. Beyond 20 seconds the stars start to get unpleasantly blurred.

And yes, I am aware that Canon is struggling a bit with ISO-invariance. My test tells that - under the right circumstances - a few EV boost isn'st o bad. To compare with another APS-C format of the time, I had downloaded and processed some Nikon D5500 shots, and what you get out of pushed up shots was really impressive. Now I know, however, that I won't get very different results by pushing +2EV my shots. Of course, I can't hope of going much farther.

By the way: your picture is amazing!!!! The landscape however was indeed much brighter than the lonely and utterly dark mountain spot where I was.

• Peter, I have applied different noise reduction settings to DPP3- and DPP4-processed pictures. I have treated the +1 EV ISO 1600 shot as if it were a native ISO 3200, and likewise the +2 EV ISO 1200 shot as if it were a native ISO 6400, each with the suggested, built-in parameters. Here's some detail:

DPP3:

=====

ISO 1600

luminance: 6

chrominance: 10

ISO 3200

luminance: 8

chrominance: 12

ISO 6400

10

14

DPP4:

=====

ISO 1600

luminance: 8.2

chrominance: 7.2

ISO 3200

luminance: 11.0

chrominance: 10.2

ISO 6400

luminance: 12.0

chrominance: 10.5

To make myself clear, for instance, the ISO 1600 +2EV and the ISO 6400 shots have been both processed with the same noise reduction amount as detailed above (DPP3 settings in DPP3, DPP4 settings in DPP4).