The Google Pixel 4 Versus the Canon RP for Night Photography – Blind Test

The Google Pixel 4 Versus the Canon RP for Night Photography – Blind Test


  1. At first the test seemed pretty easy because I was sure that A was the Canon, it was so much sharper and with better micro contrast than B. Then with increasing ISO B also started to get noisier and noisier, and this only makes sense with a traditional sensor, not with night sight. Shocking results! 😮

    Two things to note, of course: with RAW and good noise reduction (Topaz AI) the Canon would have fared much better. And more importantly, it is my understanding that Night Sight needs static scenes. Shoot moving people or concerts, and Night Sight won’t work anymore. So it’s much more specialized.

    Still, I believe technologies like this absolutely need to be introduced into traditional cameras! I would really hope for an MFT camera to get a dedicated night mode. Olympus already has the high-res mode which in some models even works handheld, and they’ve got live long exposures. Obviously, these night modes are super sophisticated, so maybe they are too high-tech for most camera companies except Sony. Would be a pity, though!

  2. The Google Pixels is just doing in camera alignment and median stacking surely?
    Was there any post processing done with the EOS RP files or were they SOOC?
    A true test of image quality would have been to also take 3-5 shots with the EOS and median stack them in post

  3. Cool video review-

    I was thinking about buying the noisy canon brick, not thinking about it anymore –

    A cheap full mirorless camera simply does not exist !

  4. holding a phone in one hand and the steering wheel with the other …while vlogging looking to the passenger seat 🤦‍♀️ … it was a nice angle to see a truck coming at you from the left at full speed!!!

  5. Now try something that move, my P30PRO also make some night mode photo better than A7iii – but for night mode you need something static for like 10 sec 😉

  6. I, too, got the pixel 4 XL and I do love it, but the colors often look flat, or "blah" compared to my Nikon DSLR/Mirrorless cameras. Then there's the "oil painting effect" on some pixel photos, which is largely inexplicable, because it happens sometimes in good light. I don't like the way the portrait mode looks. However, for nighttime shooting, it's so hard to beat and the astrophotography feature is nothing short of astonishing. If Google could get with Nikon, Canon, Fuji, or even Sony to work on color science, I think it could be a truly marvelous device, especially considering the fact that most of us have our phones with us everywhere we go.

  7. Curious to see how the XPro3’s built-in HDR works in very low light.

    Also I suspect one other downside of the Pixel 4 is capturing motion in low light, since I guess the multi-frame approach wouldn’t work there?

  8. I would have liked to have seen one shot with some movement, to really put Google's algorithm to the test. This is likely the only advantage of the Canon (although I still wouldn't expect great results!).

    I think you hit the nail on the head with your closing remarks—camera manufacturers are digging their own graves by ignoring computational photography. Aspiring photographers switch to dedicated cameras to get better image quality and performance, and then might fall in love with the handling and experience of using a camera in general despite the extra bulk and cost. But if the performance is on par (or worse), they won't ever have a reason try it out.

    Sure, Olympus has had some computational photography for a while and Fuji has been adding a bit more in recent cameras, but maybe it's too little too late?

  9. Great content! No blogger brooding over smartphones doesn't offer these very interesting comparisons. I watch you from Russia.

  10. I work underground at a limestone mine and taken many photos with my Pixel 2XL without Night Site and haven't been disappointed. I'm looking forward to upgrading to the Pixel 4.

  11. I couldn't agree more that camera manufacturer is getting lazy and smartphone is moving ahead by introducing more advance feature in computational photography. Camera manufacturer should closely look into smartphone photography feature development and bring them into their product. One thing i always dream about is to have the Lightroom preset import into the camera so i can cut down the post processing time in my workflow.

  12. Wow.. the pixel DEMOLISHED the EOS RP. I wonder how long until we get more computational photography in mirrorless cameras, we are already getting some in the Fuji Xpro 3.

  13. Pretty shocked I am. The difference in noise is quite staggering. The Pixel4 kept the shadows nicely black while the Canon washed them out in a sea of noise. Nice test.

  14. Holy Night Sight Batman!! Dayum, until now I was not buying into computational photo devices. I will definitely wait till pixel 5 rumors emerge and the price drops however, the pixel 4 is on my list. Better to have it and not need it…

  15. try ultra wide angle on both. Oops Google pixel4 doesn't have a ultra wide angle lens. And try it on a camera body with IBIS. Lens ois isn't that effective in shorter focal lens. I would like to see test on say a7iii + fe28mm or 24gm.

  16. Wow, I don't usually get things this wrong! That is really impressive for a phone. It's only a matter of time before they put this technology into cameras hopefully!!!

  17. IMHO the ghosting is from a wrongly transformed frame. It's almost mirrored around the center. Mathematically there should be more than one plausible transformation and in this case it failed to pick the one with the least difference from the rest of the frames.

  18. I picked mostly A's. Got my kid a new Pixel and kept my older Samsung. Maybe I should have given myself the new phone….

  19. This isn't really surprising. Try a stabilised lens on the Canon next time, like the 35mm 1.8 lens for the RF mount. You can get far longer shutter speeds without a tripod. Not that expensive either. Works wonders when your shooting hand held at night!

  20. It is pretty clear to notice that all the A images are flat and under poor white balance. Computer science are still far away from handling nice texture images with pleasing color rendition. However, the Canon sensor on Eos RP is so far behind in technology of noise reduction.

  21. Beautiful shots – this video is worth watching for the photos itself!
    Shocking results – had that on your channel before! Like the other people commenting.
    What I wonder, how does AI cope with moving subjects? On an old-style camera there is a choice between motion blur or going faster. I personally would also prefer some shaking to noise. Some photographers in the past were rather laid back about that. Robert Capa's D-day shots are an extreme but not singular example
    How do peopel feel abot artefacts on both devices? In your examples, I don't mind. Yet if there are humans in the frame it disturbs me.

  22. Nowadays buying expensive cameras will only be for the "potential" image quality you can achieve if your willing to put the time/effort in. Potentially with a tripod/bracketing and Lightroom you would easily beat the pixel image. And if you're a professional and get paid to do that, that's what you're going to do. But for everyone else, I think you're going to have to spend a ridiculous amount of money to match the convenience of these new phones in terms of taking decent pictures and sharing them with minimal hassle.

    Btw, I think an Panasonic or Olympus m43 with IBIS + stabilised lens would probably do a way better job. You can get 1-2 seconds shutter speed handheld with some of their models. I don't think any stabilised lens is a match for decent IBIS. However, they doesn't deal with any moving subjects well (I think pixel/iPhone can deal with that better).

  23. If only one can put the software working in these new phones behind the sensor and lenses of the actual cameras – omg! No such prospect in near future, though… And also, a 2.8 lens is slow for night photography. Get the 24mm 1.4. Even the notoriously bad low light performer RP will be significantly better with 4 times more light 😉

  24. I don't think it's a fair comparison because the cell phone computation applies things like noise reduction, hdr, sharpening, basic toning of highlights, shadows, etc. For it to be fair, I believe that you should have fully processed the Canon raw files with noise reduction, basic toning, hdr (thus the shots should have been bracketed in the Canon to be fair, etc. The Canon photos are raw files. The Google photos are highly processed photos. To me that's an unfair comparison.

  25. Personally, I don’t agree with your thoughts. From one point of view, it is excellent, the prime example of what is possible with instant heavy post-processing, but at the same time, it is a limitation. What if I don’t like heavy noise reduction with very heavy sharpening?

    So if I go out and want to take some pictures, I will grab a camera to have full control over the process. If I need it, it is easy to use multiple exposures (even handheld). Basically, I can have a picture matching my taste and have fun during a process 🙂

    Is P4 bad? No, it is the best tool if I want to share a situation instantly. 🙂

  26. Impressive. This really does not leave much territory for big sensor cameras : High resolution and low light with motion + ergonomics.

  27. The ghosting is probably due to the poor flare performance of the optics. Aside from using cheaper optics with lower quality coatings, the outermost piece of hard glass in phone cameras is almost never AR coated. This protective cover causes a lot of the ghosting/flaring in cellular photos, but is sort of needed to protect the lens assembly from dirt and mechanical damage.

  28. Thanks for making these comparison images. I totally expected that the Pixel would win this. For all of the situations where the bigger camera performs better because of the physics of light and glass, computational photography is leaping ahead in other areas. There’s no good reason that camera manufacturers cannot incorporate this tech, other than organizational hubris and denial.

  29. I hope Samsung can jump back into the camera world, if they can still profit from it. Unlike these Japanese cartels, Samsung seems to be the only one able to take advantage of computation processing and not hold back features. Just look at NX1 released in 2015 and realize how future-proof that is.

  30. The noise and sharpness threw me off completely. I thought for sure the Pixel was B due to the noise and what seemed to be lack of sharpness. When you said A was the Pixel I was shocked.

  31. You can get close to the pixel 4 result by doing the same thing it's doing.. Take a burst of a few images of the scene at appropriate shutter speed. Then stack the images as median in Photoshop or equivalent. It's of course a bit more work than just pulling out the phone. But you'll end up with more controllable result.

  32. Basically what we learn is that if you take multiple shots, you can stack them and lessen noise, and that the phone does this automatically, while it takes a bit more work with the mirrorless. The question this leaves us with is how many shots are needed for each of them, in order to obtain the required result.

  33. Surprisingly, my crop-sensor camera (panasonic gx7 mark 2) does a better job in low light than this Canon (thanks IBIS) to such extent that I don't feel the urge to buy this Pixel 4.

  34. Absolutely not surprised. Of course you get less noise if you take several photos and stack them (apart from profiled noise reduction etc.). Let's be clear: "computational photography" just means that the camera takes several pictures and combines them. So if you compare the result with a single photo, it's like apples and pears.

    @Andrew: Please do the test again but this time, take several photos with the Canon (or other "real" camera) and align and stack them. I bet you a couple of beers that the results will be the same or similar, most likely better that the Pixel's.

    Of course there is still the advantage that the Pixel does all the processing in the device automagically.

    So the one thing to take away is that camera manufacturers should have that mode to take n pictures and combine them in-camera.

  35. Well this has me blown away! The cell phone blew the Canon out of the water…sharper, brighter and no noise. Now I am thinking that AI has taken over in photography. I have spent thousands of dollars in my life on cameras, lenses, and software…computers to handle the megapixels. Now I see a cell phone that literally took better pictures (even though it was night shots only).

    And this is just the beginning of this takeover!

  36. We have to keep in mind that this pixel 4 probably takes couple of photos to produce one shot at the end also applying denoising. And here RP is taking only one photo. If you would the same process as Pixel 4 on RP and photoshop, so take couple of photos and merge them in the software then we could make some comparison.
    I would like to see Pixel 4 vs Fuji X-T2 / X-T3 JPG

  37. Great video Andrew! Did you apply noise reduction to the RP shots? Certainly you'd need a good bit of NR to compete with the phone.

    Regardless, the Pixel is super impressive!

  38. Mmmmmh people don't be foolish.
    EOS RP and all Full Frame & APS C cameras are infinitely BETTER IN LOW LIGHT than all smart device cameras from todays market even M4/3 cameras. The only reason is the bigger size on this sensors. You can make NIGHT SHOTS STACKING WITH A SIMPLE REBEL T6 FOR EXAMPLE WITH SOFTWARE AND HAVE BETTER RESULTS THAN THIS PHONE.

  39. What I think; you need to work on your photography skills. I’ve taken better photos with my canon t7i, and that’s a serious step down from an RP.

  40. This is impressive stuff from Google, but at the shot of the red neon sign, I spotted the Canon as the B camera. Now shoot something at night which moves a bit, like trees in the wind.

  41. There is one easy solution to the noise problem. Instead of editing in Lightroom , edit in DxO PL
    PRIME noise reduction is superior to nr in Lightroom.

  42. Well, them results have shocked me. Camera manufacturers are sleeping, while phone manufacturers are eating their pie. Imagine the possibilities with full frame sensor + computational photography

  43. Cool music. Sometimes the pixel was only slightly better and other times, like the chrome on the car, the Canon was better. Sometimes the lights blew our on one but not the other, and this was true f both cameras. They took turns being best.

  44. May I ask you what ISO did you use on EOS RP? (I''m sorry if you said that in the video, I wasn't able to find it) I made similar comparison between iPhone 11 Pro and Sony A7 III and the A7 III was able to destroy 11 Pro without breaking a sweat using F2.8 lens. I'm aware EOS RP uses by far the worst full frame sensor on the market that can be beaten by most APS-C sensors, but I am still surprised by how bad it is.

  45. The terrible banding in the noise immediately gave the Canon away. This is a problem that has plagued these crappy sensors since the 5DM4 days.

  46. Biggest takeaway from this: The big camera manufactures – Canikony – "are getting lazy" and loosing huge amounts of market share to the phone manufacturers, who are actually becoming the new camera manufactures, month after month.

  47. The thing that allows modern smartphones cameras to look so good is computational photography. Google have published several papers on their work at this point so if you're interested in how this all works, give them a read, they're fascinating. Smartphone cameras have tiny sensors but due to the way Night Sight works, it essentially creates a final image that similar to a large sensor, I think closer to 1" but of course you add in the stacking, noise reduction, probably some sharpening, movement detection and what it detects within a scene, that's how you end up with such stunning images. Google are doing this was the same sensor that they used on the Pixel 2 so image what they could do with something larger, more akin to what's in the P30 Pro. We've also seen that Apple really stepped up their photography game this year. You can also use Night Sight during the day too, it gives you significantly sharper images as long as you're willing to take a few seconds more for an image.

  48. I was hoping for an unbiased comparison. Unfortunately this is not such, not by a long shot (rather the exposure fall very much short of correct). Based on the metadata of the last image pair (on web page, comparison 5, a bonnet of a car), the reviewer used a 1/30-second shutter on the image stabilized Canon. That is, despite having a rated 4 stop stabilization, the reviewer chose to use a shutter speed (slightly) faster than the reciprocal rule. With the “correct” slowest shutter speed for reasonable amount of sharp images (4 stops longer than reciprocal rule is *0.6 seconds*) ISO 900 would have sufficed instead of the now ridiculous ISO 16000.

  49. This comparison is done wrong.

    If they testing with night sight then the pixel does bracketing it self as it takes multiple exposure and merge together

    The canon on the other hand was a single shot only

    So if they want to compare then, it will be single shot only not nigh sight vs single shot

  50. Of course image stacking can be done with Photoshop, etc, with any camera, including the EOS RP. But then that's some work. Also, what happens of something is moving in the image? A person walking thru for example? That makes almost any given larger sensor better than almost any given smaller sensor, as stacking can't normally be used. Also, many cameras have the stacking / noise reduction built-in.

  51. I was thinking throughout the video how smart I am and that people in the comments would be full of..
    The Pixel's cameras sucks. I mean it's nice for a device like a phone. The software is just like Adobe with google's hardware.
    The AI what is it? It's a bunch of automations. People be like Artificial Intelligence like they know what it exactly is, instead they know exactly how it exactly sounds.
    What are the automations? Noise reduction, stacking effects, and tone/contrast "correction". You could say it's actually correction but there is no right correction in our life. Heck, actually it's a permanent filter that will never give you back the raw file/s (files since it's almost always a stacking effect to be able to make the above automations to give you a single file.

    So… I've heard the news popped up on my browser. Adobe wants to make automations like that for Lightroom. Heck they might already be out but I'm not a subscriber.
    This means that it's up to devs to let your camera have exact burst modes for post production or you could even just use burst or bracketing.
    Then you go in photoshop and you can fix or achieve a lot of stuff and it will be a lot sharper and detailed.

    And all this also means that the pixel doesn't have the hardware to work on descend video quality. Video is where you see the real camera… ohh yes… you open that sucker and <joke> you can see the cosmic background radiation. </joke>

    About the ghosting of the neon? Meh, not worth figuring out. Could be a few wrong frames chose by the "AI", lens flare or w/e. It's not like you tested that angle more than once.

    1k € for that phone. I haven't bought a phone for a 7 years mine cost about 100€ back then. Thought the Pixel 4 would be a newest and greatest to upgrade but nope. I'm gonna wait for the 5. The 4 was a beta test.

  52. Hey Dontae,

    many thanks for your great video! I agree that it´s kind of shocking, that camera manufactures are sort of sleeping.

    Which light and microphone setup are you using for your videos?

  53. That's why I use a OlYMPUS camera. With it's mighty 5axis IBIS I can shot things like that @ nominal iso value. I can even go as far as 10 seconds handheld. The best IBIS in the market. Besides my camera can do focus stacking in camera too.

  54. I would assume that Canon also offers an HDR-Mode in their Mirror less Cameras, like they do in their DSLR Models. For me the test is a bit unfair. I think it is clear that a processed image and "computational photography" is exactly this, looks always better. Especially related to noise.

    So either you switch off night sight mode on the Pixel 4, or you switch on HDR Mode on the Canon.

    Otherwise, the comparison doesn't make sense to me!

  55. I have been since July take shot with Night Sight in Redmi note 7 in all shot that not action. Early google Camera mod was a lot better than stock camera. Later on new version GCam Mod has more and more mature + Option. Now I am using the url version 7.0 v12 Astro. Why not using 7.2 / later? Because it need Android 10 (the bad thing about latest Gcam in Q3 is usually need new Android version). And I am using Night Sight mode, HDR+ Enchanced, Saber, 100 Frame Number. It super sharp, super HDR, super slow, and low megapixel. Super slow here mean in 100 frame number, I can only shot once in about one minute, include background process. It just like the old days using CHDK in Canon pocket camera to save RAW. In 15 frame number, the image also super sharp, but I can shot 4x before the shutter button disable. My trick is trying to get the shot overexposure by tabing the dark color in the camera. The overexposure that seen in screen is just fake. It will create super nice HDR photo. The strange think is it can easy beat overexposure, but it cannot beat underexposure. If all dark, it will shot dark because it think that dark shot is as indended, but sometimes it bright. The detail in term megapixel also loose easly, 12 mpx (48 mpx / 4) vs 18 mpx (EOS M, my camera or in EOS RP 30 mpx). Also speed. Gcam has Top shot but it has at least use HDR+ or off (not HDR+ Enhanced).

    Please test the video but using apps Filmic Pro vs EOS RP. Which one is better? Filmic at highest res, fps, bitrate, and de-log v2 vs EOS Rp best setting.

    Thank you.

  56. Please post the exposure values for the RP shots … it seems you went a bit too far with the iso … with a fast stabilized prime like 35mm 1.8 you can shoot 1/3rd of a second handheld, and at a reasonable 12800iso with a touch of noise reduction the results can't be as bad as these you posted …

  57. Been shooting night for 6 years, I'd choose the DSLR any day. You really need to know what you're doing and you're messing with iso which is a big no no, also a tripod is absolutely critical to night photography.

  58. so is this a dishonest guy or is the tester just a ignorant reviewer…The phone is good as it stacks photos .. yes we know that …BUT SO CAN THE RP but this feature was not used …..its like testing which is the fastest a bicycle or a harley. but forgetting to switch the engine on on the harley and pronouncing the bicycle the winner

  59. With the iso jacked up so high on the Canon of course the images look this bad. This is misleading. The RP in automatic setting alone shoots way better images than anything I am seeing here. Although cell phone cameras have come a long way, they simply still cannot compete with the images that you can easily get with the Canon RP and the 24mm lens you used for this example. My thoughts are you either went out of your way to make the pixel look better or you are just not a very good photographer. Either way, this doesn't look good on you or this channel. Sorry.

  60. As a "pro" full frame shooter, I noticed from the first picture that the best was without any doubt A. Discovering that I was giving the clear win to a phone really felt like a rape, and I asked myself why the hell don't they allow computational photography also in expensive big sensors, but after all it seemed fair to me that they haven't already and probably never will. Here's why:
    even if you handhold to the point where you would use iso 25600, computational photography is not the solution a pro would use in hardly any situation. If a single shot doesn't make it for your needs, you can just take 20, auto-align and blend them with photoshop and there you have it, a quality 10 times better than computational photography but with the added freedom to edit. The point is that not even a medium format works decently at iso 51200, but what the google pixel 4 does on camera for those who don't know how to stack images, you can do it manually in photoshop if you want to choose how to make it work. Moreover, blending pictures has always a 0.1% chance of computational errors where it blends an eye with the mouth of your subject, and if pro cameras did this error in the kissing scene of a wedding (it would likely happen somewhen, if you live on shooting dozens of weddings every year), there would be no coming back, whereas with manual editing you can always return to the starting point. Definitely, computational photography is a dream for consumers, but a drawback for those who have entire days to spend doing it manually. Of course it's a different thing for prosumer cameras like the nikon d3000 and d5000 series, which will stop selling if they don't get an advanced auto mode, because in hard situations a phone can now give equal if not even better results than unedited jpegs from any camera (I cannot admit I am saying this, but the video proved it).

  61. print them and see at what size the phone camera is just pixels and the full frame would make a good sharp ish print, off course 12800 iso looks shit always.. use a tripod and shoot them at iso 100 and see the difference…

  62. I don't like to nitpick, but using an f/2.8 lens on the Canon seemed like hampering it unnecessarily. That said, the spirit of the video was good.

  63. Call me a buzzkill, but this explaining while driving shot really irks me as extremely irresponsible.
    It is one thing to talk to your audience while driving but this guy is not only looking into the camera to his right half the time but even holds the camera in his hand and it even seems that he has a script on his mobile phone in the front.
    There is no way he has adequate road awareness while holding the camera and looking to the right, and that could easily cost someone their life!

    No idea about how strict these things are in the US but in most EU Member states you aren't even allowed to hold a phone while driving.

  64. I think you didn't choose the best exposure time. Nonetheless I just compared the Night Sight mode of the Pixel 3a with the Canon M50 with stabilized 17mm ≈f/3.5 lens at 1/2s (the Pixel chose ≈1/2s, too). The results were similar, but the Pixel had a better color reproduction and handled noise better. I want to add that I also tried Canon's multi shot noise reduction (with 1/2s exposure time), but there wasn't really an advantage. I also want to add that the automatic white balance of the Pixel 3a can be much worse than the automatic white balance of the Canon M50 (when set to white priority, white priority is such a great feature of the M50)

  65. I compared the Canon M50 with the Pixel 3a again. I used the Canon M50 with a stabilized lens at 17mm ≈f/3.5 and 1s exposure time (it was so dark that the M50 needed Iso 50.000 for a bright photo) indoors. I used the Night Sight mode of the Pixel 3a and used the maximum exposure compensation (using exposure compensation in Night Sight can make a huge difference) in order to get a photo that was nearly as bright as Canon's result. The Pixel 3a combined approximately seven 0.8s exposures. I didn't use a tripod, both photos didn't show any motion blur. The Pixel 3a had better colors. The Canon suffered from more color noise and color noise reduction, Canon's auto white balance (set to white priority) wasn't really an issue, but I know that Pixel's Night Sight white balance can be often very bad. The amount of details was similar, the Canon had slightly more details in a few areas. The Pixel 3a had slightly more luminance noise (only if you use Canon's high noise reduction setting), but this wasn't as bad as Canon's color noise. I also tried Canon's multi shot noise reduction, which combined four 1s exposures. There was less color noise than before, but still more colour noise and worse color reproduction than the Pixel 3a. Some details got a little bit better with Canon's multi shot noise reduction, but some other details got slightly worse.

  66. You are using the EF 24 2.8 IS USM at the Eos RP. You chose ISO 10,000 – 1 / 60s and F2.8 for the Ford. Why don't you use the image stabilizer with 1 / 20s – ISO 2500 or 1/10s & ISO 1250? The focus is not correct, it is on the rear of the car. I don't have to see more pictures. The limiting factor is not the camera, but behind the camera.

Leave a Reply

Your email address will not be published. Required fields are marked *