{"version":"https:\/\/jsonfeed.org\/version\/1","title":"Tangram Vision Blog","home_page_url":"https:\/\/www.tangramvision.com","feed_url":"https:\/\/api.feedifyrss.com\/feed\/blog\/feed.json","description":"RSS feed for Blog","items":[{"id":"urn:sha256:80f6eb5e3a9681e2c011f8efcdb412b946fc05213f7b5ff316a9f6bb236bf349","content_html":"

MetriCal, Tangram Vision\u2019s world-class multi-modal calibration suite, has moved to credit-based self-serve pricing<\/a>. Download the software, run it locally, and pay only for successful calibrations. No more sales cycles. No forced demo calls. No more guessing whether it'll work for your sensors. Get results in hours, not weeks.<\/p>

You can start using MetriCal today. Download it<\/a>, claim your free 15-credit trial<\/a>, and run your first calibrations without entering payment information.<\/p>

How Credit-Based Pricing Works<\/h2>

The mechanics are straightforward. Every calibration costs 3 base credits, plus 1 additional credit per sensor modality beyond the first. A camera-only calibration costs 3 credits. Add LiDAR to a run, and it's 4 credits total. Camera, LiDAR, and IMU is 5 credits, etc etc.<\/p>

Here's the cool part: credits only deduct when calibrations succeed. MetriCal analyzes both your results and data quality, checking for high-risk diagnostics<\/a> that indicate problems with your data capture. If a calibration fails or throws warning flags, you don't pay. You're not burning budget while you debug bad datasets or refine your collection process.<\/p>

Credits live at the organization level, not per-user or per-machine. Your entire team can run MetriCal across multiple systems without seat restrictions. Deploy it on your development workstations, your production line, or your field test rigs; the credits pool across all of them. This removes the artificial constraints that come with traditional per-seat licensing and lets you use the software where you actually need it.<\/p>

Software That Grows With You<\/h2>

Once you've validated that MetriCal works for your sensors, choose the pricing path that fits your workflow. Credit bundles (\\$249 for 15 credits, 3-month rollover) work well for occasional needs; subscriptions (starting \\$999\/month for 75 credits, 6-month rollover, automatic overages) make sense for ongoing development. Enterprise removes credit constraints entirely for teams running high-volume production lines or needing offline deployment.<\/p>

Ultimately, you decide. There's no sales gatekeeping, no negotiation, no waiting for quotes. Pricing is transparent and published. You can evaluate MetriCal on your terms and scale up when it makes sense for your operation, not when a contract renewal comes due.<\/p>

Why We Made This Change<\/h2>

Tangram Vision has spent the last 5 years developing world-class calibration software. The technological task we face every day is daunting: create a calibration suite that is not only precise, but also works across modalities in a seamless, user-friendly way.<\/p>

The only thing harder than that was figuring out how to license it.<\/p>

Multi-modality calibration has always been a high-friction problem. You need to know if calibration software works for your specific sensor suite before committing to it, but traditional licensing forces you into a contract sight unseen. You\u2019re then locked into rigid pricing that doesn\u2019t match your actual usage.<\/p>

We learned this the hard way by being the company that forced the contract<\/em>, by putting users into those licenses. We heard the same feedback from robotics teams time and time again:<\/p>

\u201cWe need to try this in our software before we\u2019re confident\u201d<\/p>

\u201cOur calibration volume varies month-to-month; fixed pricing makes no sense\u201d<\/p>

\u201cWe don\u2019t need an Enterprise contract just to calibrate 10 systems\u201d<\/p>

Our previous licensing models (and we\u2019ve had many) either leaned too far into the development side or the scale side. We asked users to restrict their calibration modalities, limit their calibration counts, or pay for machines they didn't use. We asked small companies to accept all-or-nothing contracts that were prohibitively expensive. Nothing fit, and it frustrated everyone.<\/p>

So, over the course of 2025, we faced this problem head-on and figured out a better way. Today, that work makes its public debut.<\/p>

Get Started Today<\/h2>

MetriCal is available now with credit-based pricing. Download<\/a> the software, start your free 15-credit trial<\/a>, and run calibrations on your actual hardware. No payment information required. No sales calls. Just download and calibrate.<\/p>

See the full pricing breakdown and compare plans at tangramvision.com\/pricing<\/a>. If you're evaluating Enterprise options for high-volume production or specialized deployment needs, visit tangramvision.com\/enterprise<\/a> for details on unlimited calibrations, offline licensing, and dedicated support.<\/p>

Questions about which plan fits your workflow? Contact us directly<\/a> and chat with the experts in calibration systems.<\/p>

Professional multi-sensor calibration shouldn't require enterprise sales cycles or upfront contracts before you know it works. Credit-based, self-serve pricing puts control back in your hands.<\/p>","url":"https:\/\/www.tangramvision.com\/pay-for-what-you-calibrate-tangram-vision-introduces-credit-based-licensing","title":"Announcing Public Pricing and Credit-Based Licensing for MetriCal","summary":"MetriCal, Tangram Vision\u2019s world-class multi-modal calibration suite, has moved to credit-based self-serve pricing","date_modified":"2025-11-19T00:00:00.000Z","author":{"name":"[\"brandon-minor\"]"},"tags":["company-news"]},{"id":"urn:sha256:668c9bf506b481a0b1c0e721c3e2c8a60c8b812a1ee79fb5605a0b26895f6bcd","content_html":"

A lot of great engineers have trouble calibrating narrow field-of-view cameras, which we\u2019ll take here to be around 25\u00b0 FOV or less. Frequently, they will find that even if their calibration results in low reprojection error, the parameters will be wildly inconsistent. Repeated calibrations with many of the same type of camera and lens, or even repeated calibration of the same unit, get different answers. Even worse, the intrinsics may look reasonable, but downstream vision tasks like extrinsics calibration or stereo depth inference between cameras are now completely unreliable.<\/p>

Let's discuss the theory behind why this happens, and what we can do about it.<\/p>

Try it yourself<\/em>: You can find an example dataset and MetriCal guide for this technique in the MetriCal documentation<\/a>.<\/p>

NFOV Calibration is Tough<\/h1>

In a nutshell, it\u2019s difficult to uniquely observe the focal length (i.e. the \u201cfx, fy\u201d parameters you may be familiar with) of the camera system when using a NFOV camera. To see why, we\u2019ll briefly discuss what focal length is, how it affects image formation, as well as a quick overview of the most common calibration algorithm and the impact a narrow field of view has on it.<\/p>

What\u2019s a focal length?<\/h2>

Key to the process of calibrating camera intrinsics is estimating the camera\u2019s focal length. In intrinsics camera models, focal length is actually two quantities bound up in one value.<\/p>

In the abstract, we can derive focal length in pixels by taking the expression $\\frac{f_m}{pp}$ where $f_m$ is the focal length in meters and $pp$ is the pixel pitch in meters-per-pixel. Pixel pitch represents density of the sensor.<\/p>

The metric focal length, however, is a lot more abstract and is only physically meaningful for a real pinhole camera. Practically speaking, the metric focal length captures the \u201cwideness\u201d or \u201cnarrowness\u201d of the field-of-view of a sensor-and-lens combo. Generally a lens is thought to be what determines this quantity, but the sensor\u2019s overall size affects the metric focal length as well. This is why lens focal length is often expressed in \u201c35mm\/full frame equivalent\u201d in order to provide a frame of reference for how pictures taken with that lens look when taken with a sensor of known size.<\/p>

Taking your sensor as a given, the focal length parameter of your camera\u2019s intrinsics model primarily varies with the \u201cwideness\u201d or \u201cnarrowness\u201d of the field-of-view, which is primarily determined by the lens.<\/p>

The question for those calibrating a camera is then: \u201chow do I observe the camera\u2019s focal length?\u201d The answer is to observe focal length\u2019s unique effects on the image formation process.<\/p>

What does focal length look like?<\/h2>

When people think of camera focal length, they primarily consider the field of view and the ability to see objects that are far away. A wide field of view lens (i.e. low focal length value) lets you see objects near the camera and potentially very far off the camera\u2019s optical axis. A narrow field of view lens (i.e. high focal length value) lets you see objects far away but only within a small range around the camera\u2019s optical axis.<\/p>

Foreshortening is another related effect determined by focal length. Foreshortening describes the effect of perspective, which is the difference in apparent size of objects at varying depths. Wide FOV lenses display a lot of foreshortening. An object at a given distance will appear much larger than one even slightly further away. With a narrow FOV, foreshortening is much less pronounced. Objects set apart even a significant distance may appear approximately the same size in a narrow field of view lens.<\/p>

Consider the below video. In the scene, there\u2019s a calibration board tilted away from the camera and a few equally-sized spheres placed around it. As the video progresses, the focal length sweeps from wide to narrow. Simultaneously, the camera moves backwards to approximately maintain the size of the board in the image.<\/p>

However, nothing in this scene is actually moving. This process is called a dolly zoom<\/a>. As the focal length narrows, the scene appears to compress. Even though the closest and furthest spheres are twelve meters apart, they appear similarly sized. Also, the far edge of the board tilted away from the camera appears to pull in, with the board's apparent shape under projection going from a trapezoid to a rectangle.<\/p>

Foreshortening is very important to observing focal length; we\u2019ll see why in a bit.<\/p>

How does calibration work?<\/h2>

First, we\u2019ve got to take a brief detour to discuss how calibration works. We\u2019ve got a series of blog posts<\/a> about this if you want more detail, but we\u2019ll do a short overview here.<\/p>

In a typical calibration scenario, we take images of a target of known dimensions. The target has detectable and uniquely-identifiable features which correspond to 3D feature locations in some model coordinate system. We project the 3D features into the image using the current estimate of the camera intrinsics and find the error between the projected values and the detections. This is called reprojection error. The calibration process is an optimization problem whose objective is: \u201cadjust the intrinsics in order to minimize (the sum-squared of) reprojection error\u201d.<\/p>

In order to project a 3D point into the image, the 3D point has to be in the camera\u2019s coordinate system. The 3D points described by the target\u2019s design are in some relatively-arbitrary model coordinate system. In order for calibration to work, we also have to jointly estimate the camera-from-model transform for each image. That per-image transform is a nuisance parameter: something we have to estimate but isn\u2019t actually of interest to us in the end.<\/p>

The reason for bringing up details of the calibration algorithm is that it highlights a natural ambiguity that arises when calibration data isn\u2019t captured correctly. When a planar target\u2019s normal is aligned with the camera optical axis, focal length can freely trade off with the Z component (or depth distance) of the camera-from-model transform. Which means that in minimizing reprojection error you can choose any focal length and the camera-from-model depth (i.e. Z component) will simply compensate and vice versa. In this scenario, it is impossible to learn the focal length.<\/p>

This is why we suggest users capture a wide variety of angles between the board normal and the camera\u2019s axis. When the board is tilted away from the camera, focal length becomes observable through the foreshortening effects (like we saw in the previous video), because the board by virtue of being tilted exists at a variety of depths relative to the camera.<\/p>

The following video highlights the aforementioned ambiguity. The scene is the same as above, except the board\u2019s normal is aligned to the camera\u2019s optical axis. As before, the camera\u2019s field of view gets gradually narrower as the camera moves backwards. Again\u2014 nothing else is moving. In this scenario, it\u2019s possible to exactly preserve the appearance of the board in the image even as the camera\u2019s position and focal length change wildly.<\/p>

It\u2019s important to highlight that this isn\u2019t a binary thing. Calibration doesn\u2019t just all of a sudden work as soon as you go one degree off the optical axis. All sorts of inaccuracies introduce error into the process and reduce the observability of the focal length and thus your margin for error in the calibration data capture process. These include:<\/p>