DSLR Camera for photogrammetry. Is the D850 a viable choice?
polycounter lvl 7
Hi there!
I´ve been doing a fair bit of research the past days and reading up on many forum posts/guides on what camera one should use that is suitable for photogrammetry.
At this point, this discussion has been brought up so many times and I don´t intend to be the guy who just lines up to repeat what is already out there. However I reached a point of reading so much on it, that it would be great to hear a fresh opinion if what I concluded makes sense and that I´m not missing anything crucial before moving and buying a camera.
I put together some bullet points that I filtered from various sources:
Camera physical specs:
Looking at what many people seem to use, I boiled down to the following 3:
Entry price range: Sony A6000 (No full frame but gets mentioned many times)
Mnd price range: Nikon D750/800 /Sony Alpha A7R III
Top price range: Nikon D850 (Is 47.5 Megapixel reaching overkill number?)
Right now, I´m leaning towards the D850 as I don´t want to compromise on quality given it is a long term investment. However I´m uncertain if this camera is already way beyond the point of diminishing returns.
Thanks in advance for any input!
I´ve been doing a fair bit of research the past days and reading up on many forum posts/guides on what camera one should use that is suitable for photogrammetry.
At this point, this discussion has been brought up so many times and I don´t intend to be the guy who just lines up to repeat what is already out there. However I reached a point of reading so much on it, that it would be great to hear a fresh opinion if what I concluded makes sense and that I´m not missing anything crucial before moving and buying a camera.
I put together some bullet points that I filtered from various sources:
Camera physical specs:
- A DSLR is a must
- Full frame sensor
- GPS. Although I haven´t understood yet what for. Aerial shots?
- Avoid an electronic shutter
- High megapixels result in a higher resolution of the model and its textures. However the pixels should also be of quality.
- Exceptional low noise behavior in native to lower ISO levels.
- If the camera has noise reduction and image stabilization features, then those should be able to be turned off.
- RAW capable
- Fast autofocus to take shots faster
- A prime lens delivers the most reliable results as any zooming should generally be avoided.
- Tripod
- Remote to avoid touching the camera when taking pictures and move the object that is to be scanned.
Looking at what many people seem to use, I boiled down to the following 3:
Entry price range: Sony A6000 (No full frame but gets mentioned many times)
Mnd price range: Nikon D750/800 /Sony Alpha A7R III
Top price range: Nikon D850 (Is 47.5 Megapixel reaching overkill number?)
Right now, I´m leaning towards the D850 as I don´t want to compromise on quality given it is a long term investment. However I´m uncertain if this camera is already way beyond the point of diminishing returns.
Thanks in advance for any input!
Replies
D850 would do super well indeed although more pixels would require more time to calculate accordingly and it takes ages even with Reality Capture.
I am using Sigma dp2 quatro currently https://www.dpreview.com/reviews/image-comparison/fullscreen?attr18=daylight&attr13_0=nikon_d850&attr13_1=sony_a6000&attr13_2=sigma_sdquattroh&attr13_3=sigma_dp2q&attr15_0=raw&attr15_1=raw&attr15_2=raw&attr15_3=raw&attr16_0=100&attr16_1=100&attr16_2=100&attr16_3=100&attr126_0=1&attr126_2=1&normalization=full&widget=1&x=-0.5287724826079089&y=0.3001701393052897
and as all Sigma cameras it gives super crisp and detailed image very close to D850 doing fewer pixels at the same time. Plus it has a super crisp fixed optic.
Disadvantages are it needs an annoying 2 minutes break after each 7-9 shots and couldn't work well above 100 iso. So same problem with not enough light under too much cloudy condition. Not as a big problem as with iphone although. Doing perfectly well with just a little bit of underexposure. Got best photogrammetry with this camera than I ever managed before.
Wanted to buy Sigma H initially but was scared off by that huge heavy monster. Hate to feel all those DSLRs on my neck. I am afraid D850 would be same kind of a burden and would be collecting dust most of a time.
Big thanks for making me aware of this comparison site. I wasn´t aware something like this existed and this helps a lot. Although naturally its hard to perfectly judge depending on what settings have been chosen. Still a nice corner point to work from.
The Sigma definitely looks like a solid option however the time breaks you discribe seem quite a productivity hit, given one scan takes around 60-100 shots from what I´ve read. :O
In terms of noise, some parts of the shot that is shown on the comparison shots are very crisp however others seems fairly noisy to my untrained eye.
How was your experience in that regard? Did it perform well accross all the scans you did or were there differences between the type of objects you scanned?
As you mentioned A6000, I had pretty good results with Sony A5000 as I wanted to try photogrammetry on the cheap before I commit to anything.
It has it's limitations, but after some time, trial and error I got pretty good results (I mostly scanned trees and other forest stuff)
Tripod is definitely a must, especially if you need longer exposure time.
If you want to scan stuff like everyday objects I'd buy one of those foldable photographic tents and lights,
or make some other studio setup.
Also, think which software you're aiming for as that can be a big chunk of your photogrammetry budget,
I had best results with Reality Capture.
A DSLR isn't a bad choice, but a mirrorless camera is probably better, it will be smaller/easier to travel with and won't have a flipping mirror that can cause shutter shock.
APS-C or even M43 will be good enough for the vast majority of cases but of course FF is good too. I wouldn't put too much stock into getting a FF sensor unless you *need* 36MP or more. If you're okay with 24MP the Sony A6500 would be a very good choice, along with any number of other 24MP APS-C cameras.
The advantage something like a full frame sensor has over a smaller M43 sensor (Olympus, Panasonic) goes out the window if you're light limited and need to stop down. Let's say you have to shoot at F16 and ISO 800 on a D850 to get enough in focus. On a M43 camera you could shoot at F8 and ISO 200 to get the same DOF, while getting better dynamic range and noise performance (though at a lower resolution of course).
If you're always shooting at ISO 100 on a tripod, a FF sensor will give the best image quality though.
GPS is generally used to tag the location where you took the image. This can be nice but it's not necessary unless you have a specific use for it.
This depends highly on the camera. Some cameras can do full sensor readout with an electronic shutter, which means you will be more likely to get a crisp shot with no vibration from the shutter mechanism. Other cameras use reduced bit-depth formats for the electronic shutter, so make sure to research the specific camera model.
Right, you need both a good lens + high res sensor to get high res images. A bad lens in front of a 42MP sensor isn't going to give you a detailed image. It's worth noting that most cameras use Bayer sensors with grids of 2x2 color pixels, which needs to be demosaiced to create the final image. What this means is that with a traditional camera you'll never get 1:1 pixel detail, so if in doubt, go for a higher res sensor than you think you need.
As mentioned, some Sigma cameras use a different kind of sensor that produces more detail relative to the megapixel rating. Basically they stack the pixel sensors on top of each other so 1 pixel = 1 pixel. This generally means very good color reproduction and good detail (for the resolution), but poor low light performance. This might be fine if you only need the camera for texture reference, but I wouldn't recommend one for general photography. Sigma cameras are just, well, pretty freaking weird too.
It's worth keeping in mind you generally do not need super high resolution for photogrammetry because you will be taking many images, and the final resulting content when stitched together will be much higher resolution than a single image.
One last thing, super high res cameras that do not have low pass filters, like the D850, produce very sharp images but moire is a problem. See the DPR comparison linked above, in the fine detail there is false color from moire.
Yes, though this tends to be the case with basically any modern digital large sensor camera. You want to shoot at base ISO or very near it all the time for max detail, dynamic range and minimum noise. Exposing well is important to maintaining dynamic range and minimizing noise too. If you underexpose by two stops that's like shooting at two stops higher ISO, you'll kill the dynamic range and add a lot more noise when you bump up the exposure in post.
This depends, if the camera has IS, and you're not shooting with a tripod, it absolutely should be on. If using a tripod, it's generally a good idea to turn it off but this depends on the camera - some are smart enough to disable it automatically when mounted on a tripod. Much of what you'll read about IS systems is outdated. Modern IBIS systems in Sony and Olympus cameras are phenomenal.
Noise reduction typically only applies to JPEG, so if you're shooting RAW, which you should do, you don't have to worry about it.
Yep, and any interchangable less camera will support this.
- Fast autofocus to take shots faster
Autofocus speed is totally irrelevant for reference photography, setting up and framing the shot is where the vast majority if your time will go, focusing will happen in about a second or less. Modern cameras focus in the range of fast to very fast. Where AF systems really matter is for sports where tracking is the important thing.While you'll always get the best image quality with a camera locked on a sturdy tripod, shooting with a remote cord and a delay to eliminate any chance of shake, this can be a very slow and difficult process. If you have a camera you're comfortable shooting hand held, it will speed up your workflow significantly. If you're shooting in a studio environment and can control the lighting, it's fairly easy to shoot at fast enough shutter speeds to shoot hand held.
Lastly, research the concept of diffraction. With large sensor cameras, getting the object entirely in focus is a challenge. If it's a small object, and you're using a FF sensor, you might need to stop down to F32 to get a wide enough focus range. This means diffraction kicks in and will make your image a blurry mess. Generally, you'll hit peak sharpness at F8 on FF, F5.6 on APS-C, and F4.0 on M43.
Lastly lastly, SHOOT WITH A COLOR CHECKER and color correct from RAW in post. Your camera's white balance and exposure programs suck. It doesn't matter what camera you have, it sucks. It's not designed for color accuracy so much as perceptive color (what we perceive as appealing color), and can easily be fooled by large percentages of certain colors or different lighting temperatures.
Thanks for the feedback on the A6000. This validates my assumption with the A6000 unfortunately being a trade off in some cases. I thought of going budget first and settle later however I´m really looking into making photogrammetry a pillar of my professional work because of which I don´t feel well with compromising in equipment that I then would have to essentially re-invest in down the line.
Really like the idea of the protable studio setup. Maybe I can safe a bit there with building the tent myself. Also serves as a nice side project itself
@EarthQuake
Appreciate the in depth answers to my question!
Absolutely, I only secundarily intend to shoot free hand and primarily aim for a tripod / studio setup with a turn table to have the most control possible. The colour checker is a great point as well, completely missed this and added it to my list.
Your sentences "If you're always shooting at ISO 100 on a tripod, a FF sensor will give the best image quality though." and " if in doubt, go for a higher res sensor than you think you need." seems to make the D850 a good choice with diffraction and moire to be something I will have to learn to work with. Also considering it seems to be well rounded camera in general for when I would like it out and take landscape shots. Would you support this conclusion or advice otherwise?
Also do higher file size raws due to the large megapixel values impact the time it needs to calculate? Or is this driven by the total sum of pictures? I couldn´t find any accurate source on this unfortunately.
It's safe to assume that the size and amount of files will affect how long the scans take to process and how much memory your computer needs to have to process them.
When it comes to picking out a lens check out the DXO mark tests. They have very detailed tests with different lenses on different bodies, and you can see which lenses hold up best to the edges of the image (go to measurements then Field map), which is desirable for scans. Lenses are commonly most sharp in the center and soften up near the edges.
You can enter in various specs to the lens widget here: https://www.dxomark.com/lenses/mounted_on-Nikon_D800E-814/launched-between-1987-and-2018/focal-from-50-to-105/launch_price-from-0-to-13000-usd/lens_zoom-prime#hideAdvancedOptions=false&viewMode=list&yDataType=global
You may want a macro lens, which are generally designed for edge to edge sharpness rather than center sharpness. For Nikon maybe a 105mm 2.8 Macro, depending on how much working distance you need.
Some modern general purpose prime lenses like the Sigma 85mm and 50mm 1.4 Art lenses are very sharp to the edges too, so those are very easy to recommend other than the fact that they are quite large and heavy.
Thank you again for the very detailed feedback. Lenses will definitely the next point to tackle and this comparison site seems very useful to make an educated decision from.
I´ll soon look into this after I settled for the body. Currently I´m researching the performance hit different mega pixel sizes can make and if my hardware can take it. The 45mp of the D850 is quite massive in that regard.
I´ll share my first experiences here as soon as my setup is up and ready. Hopefully it will be of use for anyone who finds himself in the same spot.
Another Sigma disadvantage is its DNGs are super huge and require even more of those pauses. And the native RAW format is one , Lightroom can't read so you have to use pretty slow and inconvenient Sigma soft.
Right now I am looking for a good programmable photogrammetry drone rather than DSLR. We tried Mavic Pro and while it does a perfectly crisp DNG output visually, somehow the resulting geometry is infested by errors and a total mess often. My guess the image stabilization is not very good for the purpose or something.
Based on this comparison
https://www.dpreview.com/reviews/image-comparison/fullscreen?attr18=daylight&attr13_0=nikon_d850&attr13_1=nikon_d810&attr13_2=phaseone_xf100mp&attr13_3=sigma_dp2q&attr15_0=raw&attr15_1=raw&attr15_2=raw&attr15_3=raw&attr16_0=100&attr16_1=100&attr16_2=100&attr16_3=100&attr126_0=1&attr126_2=1&attr171_1=1&normalization=full&widget=1&x=-0.6233872312052007&y=0.2641721879864336
I would say Sigma may rival something around 30+
But an advantage is they do pretty good pixels/detail ratio , kind of a super crisp pixelart even in dark background of this example while Bayers always do a bit of softy interpolation in between neigboring RG and B sensors and in general more pixels to crunch for Photoscan/Reality capture.
-cheap
-5 stops IBIS (ditch your tripod)
-fold out articulating screen, take portrait orientation photos at low or high angles easily
-pixel shift hi-res mode, you'll need a tank of a tripod for this one and a static object but it is impressive.
-touchscreen, whatever, but some people like it
-did I mention it's cheap?
The sensor is trying to be stationary and the camera is moving around the sensor, which I guess could cause some extremely minor, less-than-pixel-wide-distortion depending on how rectilinear your lens is...I've never had any issues with it. At the end it mentions it's only compatible with "Image-variant camera calibration" (automatic lens distortion/focal length correction) which is the default with every photogrammetry software I've ever used, entering in manual values for focal length and sensor size has never worked for me.
The biggest issue I've had with shooting without a tripod is my own presence altering the lighting on my photoscan, usually I can take a couple steps back and use a longer focal length.
I have an a7rII and an Olympus OM-D EM5 Mark II, I use the Mark II for photogrammetry most of the time because 42MP takes four times as long to process as 16MP photos and the IBIS on the a7rII is nowhere near as capable, but the main reason is actually because of a lens, https://www.imaging-resource.com/lenses/olympus/14-42mm-f3.5-5.6-ez-ed-m.zuiko-digital/review/
with that lens it can drop into a jacket pocket and not really have to think too much about it, it doesn't get in the way and it goes back to the old saying, "the best camera is the one you have with you".
If I am not wrong Image Stabilization could make one part of an image be recorded with one lens position and other part with another one and needs a kind of mesh transformation based compensation. But how much precise could it be? How might it work with roll shutter? I think I noticed that too wide lenses are not that good for the purpose while also should be calibrated by soft to compensate it perfectly. Although it could be just a general lack of crispness rather than distortion itself.
From what I saw on Photoscan forum couple years ago it had been suggested to manually split camera calibration groups into individual groups per each image there to force Photoscan do a precise re-calibration for each next image due to IS. Personally never noticed any difference. Have no idea how it is in Reality Capture, maybe fully automatic.
I am usually just trying to avoid shooting under too dim conditions and too narrow aperture beyond f8. Beyond that It doesn't make things sharper anyway due to a light diffraction.
I recover pixel level details from the generated texture map, so I actually wouldn't have noticed if IBIS is having any detrimental effect on the point cloud, Worth testing I suppose.
It's probably there as a non-issue, IBIS wouldn't work at all if the image changed greatly from the moment the shutter opened to when it closed...it would be blurry.
On your last point, Switching focal length up mid-scan can cause issues so there's definitely some credence to what you're saying.