p.6 #1 · Star Eater Algorithm and getting in contact with Sony
snapsy wrote:
The noise targeted by this specific type of algorithm generally wont show up on quantitative noise measurements - the type of noise it targets contributes only a a tiny fraction of overall noise to the image. So it wont affect the type of measurements that DXO and dpreview perform.
You don't think that it would get the high frequency data in an MTF? I tried to interest Roger Cicala in it, as it would have been a very simple thing for him to test, but he didn't bite....
p.6 #2 · Star Eater Algorithm and getting in contact with Sony
thrice wrote:
I trust that there will be a solution, however they gave uncompressed RAW with the spatial filtering as a silent stowaway.
That's not how it happened at all - it wasn't a silient stowaway, nor was it included in the uncomressed RAW update. When the camera was released people complained about noise in long exposures so they issued a firmware update and said "this will reduce noise in long exposures", now more than a year later people are complaining there's noise reduction in long exposures...
thrice wrote:
I reckon they're 100% for sites like DXO and dpreview to compare and be like "OMG THE NOISE IS SO LOW!"
Now that a lot of those sites will be wise to this trickery they will probably roll it back as a new 'feature'. Then they'll try something else sneaky.
It was in response to customer complaints, also I don't think it would affect the DXO scores at all as I doubt they use exposures as long as 4s in testing.
p.6 #4 · Star Eater Algorithm and getting in contact with Sony
navmannz wrote:
You don't think that it would get the high frequency data in an MTF? I tried to interest Roger Cicala in it, as it would have been a very simple thing for him to test, but he didn't bite....
-John
DxO's sensor tests only measure noise, with a presumption that MTF will be equal to the sensor sampling frequency (pixel count). They do a basic data correlation test (to test for raw NR which can be used to "cheat" their noise tests), but that test wouldn't catch this. For example their test missed the fact that Sony does raw NR on the A7s at ISO 51,200. Plus DxO doesn't use long exposures for their tests.
p.6 #5 · Star Eater Algorithm and getting in contact with Sony
snapsy wrote:
DxO's sensor tests only measure noise, with a presumption that MTF will be equal to the sensor sampling frequency (pixel count). They do a basic data correlation test (to test for raw NR which can be used to "cheat" their noise tests), but that test wouldn't catch this. For example their test missed the fact that Sony does raw NR on the A7s at ISO 51,200. Plus DxO doesn't use long exposures for their tests.
But Roger Cicala's test bench would presumably pick it up because he's doing MTF based testing with a rigour that surpasses most - perhaps several of us should all email Len Rentals and ask him to do it...
p.6 #6 · Star Eater Algorithm and getting in contact with Sony
navmannz wrote:
You don't think that it would get the high frequency data in an MTF? I tried to interest Roger Cicala in it, as it would have been a very simple thing for him to test, but he didn't bite....
p.6 #8 · Star Eater Algorithm and getting in contact with Sony
navmannz wrote:
But Roger Cicala's test bench would presumably pick it up because he's doing MTF based testing with a rigour that surpasses most - perhaps several of us should all email Len Rentals and ask him to do it...
No Roger's MTF test bench would not pick this up as it tests the lens only not the camera!
Previously they have used Imatest which uses the camera body, but I think that uses a slanted line to do MTF measurements so median like filtering wont have much of an effect.
p.6 #9 · Star Eater Algorithm and getting in contact with Sony
Matt Grum wrote:
No Roger's MTF test bench would not pick this up as it tests the lens only not the camera!
Previously they have used Imatest which uses the camera body, but I think that uses a slanted line to do MTF measurements so median like filtering wont have much of an effect.
p.6 #11 · Star Eater Algorithm and getting in contact with Sony
Thank you! Now I finally see similar output as the posted earlier and the case is solved. 42MP state-of-art is turned into cellphone quality right there. Nokia 808 output is actually better in good light.
p.6 #12 · Star Eater Algorithm and getting in contact with Sony
navmannz wrote:
But Roger Cicala's test bench would presumably pick it up because he's doing MTF based testing with a rigour that surpasses most - perhaps several of us should all email Len Rentals and ask him to do it...
-John
The bench tests just the lens - we eliminate the camera and test just the optics - so it wouldn't help here.
p.6 #16 · Star Eater Algorithm and getting in contact with Sony
to illustrate a bit differently the effect of star eater than the 3.2s/4s comparison, here is a comparison between:
- my $3.5k Sony A7s2
- my first digital camera, the Canon 350D / Rebel XT. the camera body is 11 years old, and cost now $150
- with the same lens, same aperture, same ISO, same everything
with its 6µm pixels, 350D is more resolve, shaper, and infinitely more sensitive in terms of detection of weak stars than my sony a7s2 with its 8µm pixels and star eater!
guys, do not hesitate to post this comparison elsewhere on the web to make sony react
Maybe it's indeed a time to do a massive social media blow?!
...hit sony twitter and facebook sites and post all these links. It did work with uncompressed RAW case some time ago!
p.6 #19 · Star Eater Algorithm and getting in contact with Sony
I did some testing last night on an A7rii firmware 3.10 on a couple different lens, 3.2 sec. vs. 4.0 sec both not bulb compared. As well as longer times (over 30 sec. tracked) to compare bulb mode
