What kind of accuracy can I expect from Canvas?

The Short Answer

If you're using Canvas LiDAR 3D Measurements on a compatible LiDAR-enabled iPad or iPhone (includes most Pro models launched since 2020), most measurements in your CAD file should be within 1-2% of what's verified manually with a tape measure, laser distance meter, or existing blueprint. 

If you are using Canvas Lite for Homeowners without a LiDAR enabled iOS device, most measurements are likely to be within ~5% of what's manually verified via similar methods, but you may find some measurements outside of that range. You can learn more about using Canvas without a 3D sensor here: Can I use Canvas on an iPad or iPhone without LiDAR?

You can learn more about compatible devices here: What device should I use with Canvas?

The Long Answer

The accuracy of mobile 3D scanning depends a lot on the user and the environment, so unfortunately an absolute tolerance, e.g., "all measurements ever will always be within X millimeters," can't be provided. A football field will have more absolute error than a kitchen, as will a hall of mirrors.

Overall, we tend to hear from our customers that Canvas is an effective, powerful, and out-of-the-box solution for things like design, planning, pricing, estimation, sales, or documentation conditions (at the beginning of the project or as an ongoing thing). 

If you're looking to use Canvas for installation and construction projects that require extremely tight tolerances, you may want to verify the few critical measurements you need manually, and then adjust your CAD model once you receive it. We find that this typically still creates a large ROI, as most projects only have a few measurements that really need to be millimeter-level precise, and Canvas still automates all the other measurements and creates the 3D model.

So, what can I expect in terms of hard numbers?

Across the case studies we've run with professionals out in the field — using Canvas on real-world projects — we see that most measurements in the final CAD file generated are within 1-2% of what's verified manually by tape measure, laser range-finder, or existing blueprint. These case studies have been conducted using Canvas with a LiDAR-enabled iOS device.

Canvas Lite for Homeowners can also be used with an iPhone 8 or higher with no 3D sensor. But, without a 3D sensor, most measurements are likely to be around 5% of what's manually verified via such methods, but some measurements may fall outside of this range. We only recommend using Canvas Lite for initial trial purposes (i.e., before you commit to purchasing a LiDAR-enabled iPad or iPhone). We do not recommend it for professional use. You can learn more about using Canvas Lite without LiDAR here.

Some of the dimensions in my final CAD model exceed the expected variance. Why?

Normally, accuracy tolerances outside of this range are simply because we received a poor scan, which in turn was because of poor scanning technique (like zig-zagging around the room), poor calibration, or other first-time user mistakes we call out in our scanning tutorial video. We always recommend scanning a few rooms around your own house to understand how Canvas handles different kinds of environments before you take it on a job site. 

We also understand that when you're totally new to 3D scanning, it can be difficult to understand the difference between "good scan" and "bad scan," let alone what may be causing it. For this reason, we'd be happy to review your first scan to help provide feedback on scan technique or anything else that will help you be successful out in the field. You can reach out to us at hello@canvas.io to get that process started.

Depending on the scan, we may need to make assumptions where the 3D sensor on your device does not directly observe the dimension in question. For example, if curtains cover a window, Canvas will typically interpret that there is indeed a window behind the curtains, but the curtains may cover the outer dimensions, so its placement within the wall may be less accurate. An even more common example is wall, floor, and ceiling thickness: Canvas does not observe these directly, so we use the scand data and construction standards to determine the dimension . A 1" difference in wall thickness accuracy could add up to be a lot more across the length of an entire building. 

You may also find that smaller dimensions (such as those under one foot) may have slightly wider variances than 1-2% because even being 1/16th of an inch off may be larger than 2%, and sub-inch precision is simply going to be hard to maintain consistently through the simplification process of Canvas. For example, wall thickness may come in at 4.5", but in reality, it is 4.625" — this is only a 1/8th of an inch difference, but it is a 3% error. 

Some kinds of environments can be a little harder to scan in general — such as a completely empty space with no furnishing and nothing on the wall (i.e., nothing for the sensor to track against), or a room with tons of very large mirrors — and may lead to slightly wider tolerances. We recommend reading:

Finally, it is important to note that humans can make mistakes too! We have debugged many orders where someone was certain that our measurements were inaccurate, but it turns out that someone had their laser range-finder against the wrong surface, or made a typo, or some other issue that had nothing to do with the scans. 

If you have questions or concerns about a recent Canvas order, please email support@canvas.io.

What if I need better accuracy?

If you need tighter tolerances than 1-2%, you can also submit manually verified dimensions for key measurements, and we will "override" the scan's dimensions with your own, and you will receive a CAD model that uses the dimensions you provided us. This is a great option if you need to get a 3D model of an entire room or home, but only a few dimensions are critical. 

To learn more about how to submit manual dimensions with your order, please read: I need higher accuracy than 1-2%. Can I submit any measurements of my own of my own for my order?