Classic stereoscopic photography is based on two images taken from slightly different viewpoints. Canon proposes a different approach: using additional information recorded when taking a single photograph to automatically generate a 3D model. Dual Pixel 3D technology, presented at CP+ 2026, is intended to use solutions derived from Dual Pixel CMOS AF.
This is an important distinction. It is not about simply splitting a single photograph into artificially shifted views or recording an image with a special stereoscopic lens. The starting point is intended to be a single exposure captured with a standard camera and lens and then processed by software. However, publicly available materials do not describe every stage of this process, so the demonstrated capabilities should be distinguished from a technical interpretation of how it works.
What Canon demonstrated at CP+ 2026
According to materials prepared by Canon for CP+ 2026, the Dual Pixel 3D system can automatically create a high-resolution 3D model from a single photograph. The manufacturer links the software to Dual Pixel CMOS AF technology, known primarily as a focusing solution.
An independent report from the trade show also described a demonstration of creating a 3D image from an ordinary photograph without using a dedicated VR lens. Meanwhile, Canon's corporate publication for 2026–2027 places Dual Pixel 3D within the broader field of spatial imaging, aimed at digitising and visualising spatial information.
The verified scope of the demonstration can therefore be summarised in three points:
- the input material is a single photograph,
- the software uses technology derived from Dual Pixel CMOS AF,
- the result presented by the manufacturer is a high-resolution 3D model.
However, the available information does not make it possible to determine the full list of compatible cameras and lenses, output formats, computational requirements or the degree of manual control over the reconstruction. It is also unknown in what form the technology will be made available to users. A demonstration of capabilities is not the same as a fully documented, production-ready workflow.
How a single photograph can contain depth information
An ordinary flat photograph records the distribution of colours and brightness within the frame. It can be used to identify perspective, object occlusion or changes in apparent size, but it does not directly provide the two views required for a classic stereogram. Dual Pixel 3D is also intended to use phase information recorded at pixel level and associated with the operation of the focusing system.
Technical interpretation: phase data can be treated as a source of information about small differences in the positions of image details. If the software can relate these differences to the positions of surfaces relative to the camera, it can estimate depth for individual areas of the frame. This produces a depth map, which is a representation that assigns relative distances to parts of the image.
A depth map is not yet a complete model of an object. Rather, it forms a bridge between a photograph and a spatial representation of the scene. The software must combine the image data with the estimated geometry, identify surface boundaries and decide how to represent areas that the camera did not see directly.
Canon has not disclosed the detailed reconstruction algorithm in the materials provided. It should therefore not be assumed that the process is based exclusively on one type of data, nor should specific computational methods be attributed to it. The only certainties are the stated link to Dual Pixel CMOS AF and the result in the form of a 3D model created from a single shot.
Dual Pixel 3D versus a classic stereoscopic pair
In stereoscopic photography, two images correspond to the views of the left and right eyes. The differences between them, known as parallax or disparity, allow the viewer to perceive depth. Such a pair can be displayed side by side, converted into an anaglyph or directed separately to each eye using an appropriate projection system or display.
Dual Pixel 3D changes the order of this process. Instead of first recording two complete views, the system starts with a single photograph containing additional phase-related data. The software uses this to reconstruct a spatial representation of the scene. Only then can such a model potentially serve as the basis for various forms of visualisation.
Two views versus an intermediate model
In classic stereo, the source material is a ready-made pair of perspectives. The photographer must control the relative positions of the cameras or the successive positions of a single camera. Depth remains encoded in the differences between the images.
In the approach presented by Canon, the central stage is the reconstruction of a 3D model. This means moving from a photograph to a description of space, rather than merely preparing a second image for the opposite eye. This distinction matters because a model and a stereogram are not the same type of data.
A stereogram makes it possible to see a scene from a specific pair of viewpoints. A 3D model can be used to generate different perspectives, but their quality depends on the completeness and accuracy of the reconstructed geometry. Canon's materials confirm the creation of a model but do not specify in detail all the possible ways of presenting it.
A single exposure as a practical difference
Capturing a single photograph eliminates the need to make two separate exposures at the time of shooting. This may be important when the scene is changing and two shots taken in succession would not be identical. However, it does not automatically mean that every scene will be reconstructed flawlessly. The quality of the result depends on how much reliable spatial information can be derived from the source data and how the software handles ambiguous areas.
What a single shot cannot see
Every method starting from a single camera position faces a fundamental limitation: a photograph does not contain a direct image of surfaces concealed by other objects. When the virtual viewpoint is changed, areas may appear that were not present in the source shot. They must remain invisible, be simplified or be filled in computationally.
This limitation distinguishes reconstruction from the actual recording of a scene from multiple directions. A stereoscopic pair does not show the entire object either, but it contains two genuinely recorded views. A system based on a single exposure has only one primary direction of observation, supplemented by additional phase information.
Difficulties may also arise at object boundaries and in areas with ambiguous structure. However, these are general consequences of reconstruction from a single viewpoint, not published Dual Pixel 3D test results. Without detailed samples and a description of the algorithm, it is not possible to reliably assess how Canon's system handles particular types of scenes.
A 3D model does not automatically mean a stereo image
Materials concerning Dual Pixel 3D use the terms 3D image and 3D model. In the context of stereoscopy, precision is important: a spatial model describes geometry, whereas a stereoscopic image is a pair of projections intended for the left and right eyes.
For a model to become a stereogram, at least two appropriately positioned views must be generated from it. They can then be saved or displayed using the chosen presentation system. However, the information provided does not specify whether Canon's software offers the export of stereoscopic pairs, depth maps or specific formats intended for 3D devices.
It is safest to regard Dual Pixel 3D as a spatial reconstruction technology that may shift the boundary between ordinary photography and 3D material. Its potential use for stereoscopy is a logical possibility arising from the nature of the model, but it should not be equated with a confirmed feature of a specific product.
Potential significance for spatial photography
If the process under development retains the stated simplicity of a single shot, it could integrate the capture of spatial information into work with a standard camera and lens. This would represent a different philosophy from using two cameras, a special VR lens or a dedicated stereoscopic rig.
The significance of this concept extends beyond merely achieving a depth effect. A spatial representation of a scene can serve as a starting point for visualisation from different perspectives. At present, however, there are no grounds for attributing specific production applications, accuracy parameters or an advantage over specialised capture methods to Dual Pixel 3D.
Assessing the technology will depend on answers to the questions left open by the demonstration: what scenes can be reconstructed, how much freedom there is to change the viewpoint, which elements of the process are automatic and how the model handles areas not visible in the photograph. Export formats and the ability to use the result in existing spatial imaging systems will also be important.
What is already known and what remains to be seen
It is a fact that Canon publicly presented Dual Pixel 3D at CP+ 2026. The manufacturer states that it can automatically generate a high-resolution 3D model from a single photograph using software derived from Dual Pixel CMOS AF technology. An independent trade-show report confirms a demonstration of creating a 3D image without a dedicated VR lens.
The precise way in which phase data is converted into geometry remains a matter of interpretation, as does the extent to which the result can be used to create stereoscopic pairs and other forms of spatial presentation. The available materials also do not make it possible to assess the accuracy, scene-related limitations or maturity of the overall process.
Dual Pixel 3D should therefore be viewed not as a replacement for classic stereoscopy, but as a different way of acquiring spatial data. A traditional pair records two real views, whereas Canon's solution is intended to reconstruct a model from a single shot and phase information. It is a promising direction in computational photography, but its practical value can only be assessed once the full process and the capabilities of the resulting model have been disclosed.
