Or: How I grew tired waiting for Insta360 to explain why their “6.4K” stereo camera can only produces relatively sub-par 4K footage.
Resolution: The degree of fineness with which an image can be recorded or produced. The amount of detail an imaging system is able to resolve.
Resolution is often used interchangeably with pixel count, even though they signify two very different things. This is understandable, since in an imaging system (like a camera), ideally the two should be equal. Optimally, the amount of detail an imaging sensor can record (the resolution) should be encoded into exactly the number of pixels (the pixel count) required to represent that amount of detail.
Update 26/06/2018: Insta360 has commented that they are working on a firmware that improves on this issue, and that it is due for release in june or july. I will update this post when it is released.
If the imaging system can resolve more detail than is actually recorded in the output data, the system is said to be supersampling (also called downscaling). Conversely, outputting to a larger pixel count than was actually resolved by the sensor in the first place is called subsampling (or upscaling). Upscaling will never produce more resolution in an image. It will only stretch the information that was already there over more pixels, increasing the pixel count.
I think it is reasonable to expect camera manufacturers to be talking about the resolution of their cameras when publishing specifications, not mere pixel counts. I find this should be true for any reputable manufacturer, but especially for companies who claim to provide equipment to professionals.
If a manufacturer was found to be blatantly misinforming about actual resolution, and just specifying arbitrary pixel counts for their product, it would surely be a great cause for concern for anyone thinking about purchasing products from them. Especially products with a “Pro” price tag.
Searching For Some Clarity
Insta360 claims that their Insta360 Pro camera is capable of producing 6K stereoscopic 360 video, or more precisely a stitched resolution of 6400×3200 per eye. To produce this, each lens needs to capture at least 3200 points of vertical resolution. Ideally a little bit more due to the distortions needed when stitching, but let’s forget about that for a while, and set the bar at 3200 points.
Looking at the recorded video files from a single lens, the pixel count matches: 3200×2400 pixels (the sensors are rotated 90 degrees, so the vertical axis in the real world becomes the horizontal axis in the recorded file). So far so good, but are the pixels actually filled with data captured at 3200 points of resolution?
Answering this question is difficult without knowing the exact image readout mode of the sensor, the processing path, encoding steps, and so on. But fortunately there’s a relatively simple trick to estimate it quite accurately, without spending a month reverse-engineering the camera.
If we take a frame of video produced by the camera at 3200×2400 pixels, and then downscale it by some factor, and subsequently upscale it to it’s original pixel count, we should be able to see a loss of information. For simplicity I will flip the axis to the file-native notation for now, so let’s designate the original pixel count of the camera’s “6K” mode as 2400p.
If for example we downscale the file to 50% of it’s original pixel count at 2400p, we will destroy any information that required a capture resolution of more than 1200p. This should significantly reduce the amount of detail in the image. It turns out that it does, but only very slightly.
After resampling video frames from Insta360 Pro’s “6K” mode to various pixel counts, it’s become clear that you can subsample the frames down to around 1500p without any loss in resolution.
Here’s two zooms of TIFF images produced from the same video frame. Click the images to download full-size TIFFs.
The first is a direct grab of the frame at 2400p:
The next one has been subsampled to 1500p, and then upscaled back to 2400p with a bit of sharpening applied:
Can you spot the loss in resolution? Neither can I.
The resolution of these files are nowhere close to the advertised resolution. In fact, I’d venture a guess that the current “6K” recording mode is actually sampling each sensor at 1536p (or possible 1792p), upscaling to 2400p, and then applying a bit of sharpening. It’s a guess, but I’d be surprised if it’s too far from the truth.
In conclusion, the Insta360 Pro does not in any way record “6K” stereo footage. The maximum effective resolution I would attribute to it is somewhere around 4K, and that’s being overly generous.
Aditionally, the “5K” mode records files at a 2560×1440 pixel count, but the resolution in those are similarly just 1080p, upscaled to 1440p.
Why Insta360 choose to do this is anyone’s guess at this point. Maybe it was a deliberate marketing ploy. 6K does sound a lot more attractive than 3K. Maybe someone was not able to deliver the actual performance they thought they could pull out of the camera, and chose to disguise it instead of admitting to the error. Maybe it’s something completely different. Maybe we will never know.
Until Insta360 comes forward with some answers, I can only recommended to be very cautious, and not take specifications from Insta360 at any kind of face value.
Waiting in Vain
I have repeatedly asked Insta360 about the issue, both in their public Facebook groups, and privately over email and instant messages. Every time it has been more or less shrugged off with a “we’ll look into it”, followed by complete silence. It’s been 10 months now since I first brought it up with them, and I have grown tired of waiting in good faith. I really still hope the camera is fixable, and that I didn’t pay 4.000 € for a sub-par 4K camera.