Best DSLR or ILC for Night Sky Imaging?

Started Nov 8, 2017 | Questions thread
rnclark Senior Member • Posts: 3,715
Re: dpreview comparisons: Are they kidding?

kiwi2 wrote:

rnclark wrote:

Just to be clear, the dpreview comparison above is more about the changing lenses used than the sensor. DPreview changes focal length while keeping f-ratio constant. That means lens aperture area changes between cameras, thus the light delivered to the sensor is changing, and that means the amount of photon shot noise is different due to the lens, not the sensor.

Download the images and look at the exif to see what focal lengths and f-ratios were used. The Sony 7RII is full frame, so the use the longest focal length and largest aperture area.

Well the 100D was 50mm f/5.6 = 8.9mm aperture

7Rll was 55mm f/5.6 = 9.8mm aperture

X-T2 was 56mm f/5.6 = 10mm aperture

Maybe look again. I downloaded the raw and jpeg files for the dpreview comparisons and extracted data directly from the files. Here are the results:

12800__1/5000____f/5.6___85.0____191_____Canon_EOS_5D_Mark_II__EF 85mm_f/1.8

So not only are the lens collections areas all over the place, but the exposure times vary too. That means the light delivered to the sensors are NOT the same. There is a factor of 19.3 in aperture areas (thus the mount of light that can be collected per unit time), and a factor of 2.5 in exposure time.

Then I checked the Sony a6000 and Canon 5d2 as they seemed surprisingly noisy. I brought the raw files into photoshop and with my default settings (and this means no vibrance, no clarity, no adjustments for exposure, dark, highlights, etc. Just some nominal noise reduction (documented on my web site) - nothing strong and most of the settings are the default settings when I first installed photoshop. This is what I get:

dpreview raw results are so bad I couldn't make my conversion as bad when I tried. Middle is dpreview comparison web page, left is full resolution CS6 ACR conversion, on the right is theconversion from the left enlarged to match the sizes in the dpreview web site.

Shocked, I'm shocked and stunned that the dpreview results are that bad. I would bet my 5-year old granddaughters could do better than dpreview with the raw converter. This illustrates what I've often said, that images like these are comparison of raw converter algorithms and not about the sensors. My already low view of these kinds of comparisons just sank through the floor. It was quite trivial to produce a far better image than the dpreview comparison page. I couldn't try the 80D or Fuji because I only have CS6 and ACR in CS6 won't open them.

dpreview web page:

BUT WAIT. That is not all. The gains (ISO levels) are set differently with each camera by the manufacturer, even within each manufacturer. So the same ISO DOES NOT MEAN THE SAME SIGNAL. For example, the 5D2 at ISO 12800 has a range of 0 to 460 electrons. An 80D has a range of about 0 to 230 electrons. So the same magnitude noise in the 80D image means a factor of two smaller noise in electrons.

To compare images properly, the images need to be calibrated to the same signal level. That means same incident light level onto the sensor, and the output comparison images MUST BE on the same scale. This is not done in the dpreview site nor on most other review sites.

Your other posted image that you said was not 14 mm f/5: the exif data shown says 14 mm f/5:

14 mm f/5 in the exit data

So I tried a bit of an experiment tonight with the same ISO and exposure time and focal length and f-ratio/aperture between the Fuji and Canon. (35mm f/4.5, ISO 6400) Partly cloudy and was shooting at a few gaps between the clouds. Pretty awful night really with poor seeing. But just wanted to see if noise would be the same according to your it's only aperture that counts theory.

Just 100% crops of the camera produced jpgs...

Is this a raw conversion using dpreviews settings on the canon image? As I show above, the color splotches originate in the raw converter, not the sensor. Better settings like I show above would make a far better image. But even with the poor conversion, the canon image shows fainter stars than the fuji. It looks like the fuji image is heavily filtered, either in the raw data in the camera, or in the raw converter. So there is no realistic comparison here. If the canon image were converted better, the canon would look better making the two images much closer, as one would expect from real sensor data.

Here is the deal. All recent sensors from all manufacturers have reasonably high quantum efficiency, typically around 60% +/- 10%. And read noise at high ISO is pretty small, around 2 electrons or so. At these levels there is very little difference between sensors for short exposure of a minute or less in real world imaging. The technology is pretty mature. Sure there is room for improvement, but it is quite small in he scheme of things. The differences between cameras seen on review sites from dxo, dpreview and others is mainly dues to varying light levels delivered to the sensors, and now it is obvious that raw conversion algorithms and basic settings move things all over the place. That leaves open the problem of a biased review, intentional or not, by some poor settings in a raw converter.

Again, the differences in making a nightscape image, the main topic here, has more to do with the lens collecting the light, not the sensor.

Analogy: Internet discussions about camera sensors are like people standing around holding buckets and claiming how big and shiny their buckets are, and not about the quality and amount of water they fill their buckets with, when the water is what they drink. "Hey look: the water in your bucked has green, red and brown splotchy things floating in it." "Oh that doesn't matter because my bucket is bigger and shinier than yours!"

Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow