Canon C70 - Testing Highlight Roll-Off

For months, I have been hearing about how much worse the highlight rolloff on the Canon C70 and C300iii are when compared to the C200. Here are a few quick test shots. Both cameras are at ISO 800, matched settings, same lens. I exposed the C70 using the zebras set at 85%, which seems to be the clipping point in CLog2, and then matched the settings on the C200 (which might actually be clipping at around 83%).

C70 @ f5.6 - Center of lightbulb highlight is just barely clipping.

C200 @ f5.6

Now let’s add the same C-Log2 to Rec709 LUT to the footage, and try bracketing a stop over this exposure and a stop under to really wrap our heads around the characteristics of the rolloff:

C70 @ f5.6 - Center of lightbulb highlight is just barely clipping.

C70 @ f4

C70 @ f8

C200 @ f5.6 - Center of lightbulb highlight is just barely clipping.

C200 @ f4

C200 @ f8

Based on this test, I’d say that these cameras have almost identical rolloff characteristics. If I really squint and stare at it, it would actually be the C70. It looks just slightly smoother to my eye. But only just barely.

So, why is everyone convinced that Canon’s new DGO sensor isn’t as good at rendering highlights? My theory is that some DPs aren’t clipping their highlights at 85 IRE, either by using the zebras or the nifty new false color feature (it looks to be very useful!).

On the other hand, the C200 shows a waveform that reflects C-Log3 rather than C-Log2 when recording in Canon Raw Lite. From personal experience, I know that I can put the highlights right at the top of the C200’s waveform and still have an extra stop of headroom. Anyone who isn’t doing this is probably protecting their highlights by a full stop, so it makes sense that C200 footage would be a lot less likely to be clipped than C70 footage.

Next, I’ll take a look at autofocus.

Canon C70 - Testing ISO & Noise

I finally have the C70 in my hands. It’s the first time I’ve preordered a camera. I did it because as good as the results as I’m getting from the C200, the workflow with Canon Raw Lite is tying up my workstation for longer than I would like. In theory, the DGO sensor in the C70 offers a faster path to a clean 10-bit 4:2:2 output for my clients.

Before I bring any camera on a job, I need to have an understanding of what it can and can’t do. There is already plenty written about the handling and physical controls of the camera. What will follow will be a few tests and observations of the All-I 10-Bit footage from the C70, using Canon Raw Lite on the C200 as a familiar baseline. Other people may approach this differently and these are mostly for my own reference, but I am posting them here in case others find them useful.

First things first. Let’s start off with a simple body cap test to get a sense of noise structure within the cameras.

Based on this test, I'd rate the C70 as usable and clean for professional work at ISO 1600, which is the same as XF-AVC on the C200 (from memory, not tested here). For Canon Raw Lite on the C200, I'd say ISO 400 is about as high as you'd want to go unless you aren’t shooting any shadows under 20 IRE or are budgeting in time to run a noise reduction pass.

However, the amount of noise in the blacks is only part of the story. Things look a little different in a controlled scene with a LUT involved and some cameras fall apart at different points in the waveform. Here are both cameras, same settings, same lens, same LUT, and I’ve reframed the shots slightly to match:

Viewing this scene on a waveform side by side at ISO 800, the C200 exhibits some striations that the C70 does not:

C200 - Note the horizontal lines in the waveform.

C70 - No lines.

I’m no colorist and I don’t know what’s causing this, but it reminds me of what the banding that appears in blue skies looks like on a waveform. Sometimes you can see an image breaking apart in the scopes before you can in the image and I wonder if that’s what is happening here. In any case, I think the C70’s signal looks like a healthier starting point for an aggressive grade.

A few other takeaways:

• Without counting on time for noise reduction, I’d use the C200 at ISO 400 and the C70 at ISO 800 (or potentially 1600 in a pinch).

• The C70 is just barely sharper.

• The C200 is a little warmer and less magenta. I prefer the look of the C200 but I can easily achieve this with the C70 in post, or I may dial in the colors in-camera.

• The C70 has a lot less chroma noise than the C200.

• It took me 49 seconds to export the C70 clips from Premiere on 2017 iMac Pro, and 2:10 to export the C200 clips. On longer projects, that will result in a significant time savings.

Next I’ll be testing dynamic range and highlight rolloff.

Denoising Canon Raw Light with Neat Video - Export Times

The Canon C200 is a great camera, especially shooting in Canon Raw Light. However, it can have a noisy image at the native ISO of 800 and above. I often run my footage through Neat Video, which does a great job of cleaning up the noise. Unfortunately, it takes a few hours and is pretty resource-intensive so I can’t really do any editing work while I’m running these exports, even if I export with Adobe Media Encoder.

I wanted to see if I could improve the export times, so I made all of the tweaks recommended by NeatVideo. I then recorded a 1-minute clip at ISO 800 and applied both NR noise reduction and a standard Rec709 LUT to it in Premiere CC 2020 before exporting.

Here are the specs of my main editing workstation:

And the results.

Footage on connected MSATA drive (SolidPod, straight from the camera)
Noise Reduction - Alexa LUT - Export to ProResHQ on Desktop
7:50

Footage on connected MSATA drive (SolidPod, straight from the camera)
NO Noise Reduction - Alexa LUT - Export to ProResHQ on Desktop
2:49

Footage on OWC SSD
Noise Reduction - Alexa LUT - Export to ProResHQ on OWC SSD
7:12

Footage on OWC SSD
Noise Reduction (after optimizing with Neat Video tool) - Alexa LUT - Export to ProResHQ on OWC SSD
5:55

Footage on Desktop
Noise Reduction (after optimizing with Neat Video tool) - Alexa LUT - Export to ProResHQ on Desktop
5:18

Footage on Desktop
Noise Reduction (after optimizing with Neat Video tool) - Alexa LUT - Export to ProRes422 on Desktop
5:18

I also tried it on my PC laptop (MSI Predator with GTX 1060):

Footage on connected MSATA drive
Noise Reduction (after optimizing with Neat Video tool) - Alexa LUT - Export to ProResHQ Desktop
16:39

And finally, I did a test with some footage from the new Canon C70, which may have a clean enough image not to warrant noise reduction. But I did have a noisier nighttime shot at ISO 1600 or 3200

Footage on Desktop
Noise Reduction (after optimizing with Neat Video tool) - Alexa LUT - Export to ProResHQ on Desktop
5:54

---

Conclusions: The Neat Video Performance tool made a bigger difference than anything else. Also, working locally and not from an external drive also helped. Exporting to ProRes422 didn’t have much benefit over ProResHQ in terms of render time. And denoising the C70 footage took longer than the Canon Raw Light footage from the C200 (though it does appear much cleaner at ISO 800).

And the award for most pleasing grain structure goes to...

I shot some comparison footage on a rainy day. I used the same lighting for each shot. All cameras were set to ISO 800 and were fitted with the same 50mm Canon lens (except for the Red, which had a 50mm Nikkor AIS). I added a noise-popping curves adjustment.

Red One MX - Redcode 36, 4K HD, Red’s Chromatic LUT

Canon C200 - Canon Raw Light, C-Log2 with a Rec709 LUT

Canon EOS-R 4K - Internal 8-Bit C-Log, C-Log to Rec709 LUT

Canon 5Diii with Magic Lantern Raw - No adjustments, just enlarged from 1080P to 4K and converted to ProRes4444

My big discovery is that the 9-year-old Red One MX has the most pleasing noise structure of any of my cameras. This old sensor continues to surprise me.