Overview
Hailstorms damage about three million cars around the world each year. If car owners could easily and reliably measure their hail-damaged cars in their own driveways, the global auto insurance/repair industry would be transformed. We are excited to show you how Quidient Reality® can turn a casual iPhone video into a relightable millimeter-accurate 5D reconstruction of a hail-damaged car. We capture videos of our Fiat in our imaging lab at Quidient Technology Center and show reconstructions of the roof and hood. The results of this quality are out of reach for current scene reconstruction products and modern physically based inverse rendering systems. Typically, only specialized laser scanners can produce high-quality results on such scenes, and only after the car is sprayed with a coating. But we show that our smartphone-captured reconstructions compare favorably to laser scanner devices at a fraction of the cost and time.
Reconstruction Pipeline
We begin by capturing one video for the roof and another for the hood. For both videos, we use the Blackmagic camera app on an iPhone 16 Pro Max to demonstrate that we do not need specialized cameras, engineered lighting, or active imaging systems. Our reconstruction algorithm does not require pre-calibrated images or video as input, which distinguishes it from most physically based inverse rendering systems today that require precisely pre-calibrated linear RAW images. No pose estimation aids, such as ArUco markers, are used for these captures; the intrinsic and extrinsic camera parameters are estimated from scratch. Videos 1 and 2 show the captured videos used for reconstruction. We capture the environment towards the end of each video, as this helps Quidient Reality® perform Generalized Scene Reconstruction.
Video 1. Fiat roof capture
Video 2. Fiat hood capture
After the videos are captured, we upload them to our desktop equipped with a NVIDIA RTX 4090 GPU running our Quidient Reality® API. Quidient Reality® can use Generalized Scene Reconstruction to disentangle the matter and light field to extremely detailed levels. Figure 1 shows qualitative and quantitative comparisons of our geometric mesh reconstruction compared to a Surphaser 75 Ultra Short Range laser scanner ground truth. The car was coated before scanning with a commercially available scanning spray. Our reconstruction was manually aligned to the laser scan with Cloud Compare.

Figure 1. Geometric reconstruction comparisons against the Surphaser 75 USR structured light scanner as Ground Truth.
Relightability
Quidient Reality® is also capable of material reconstruction using Generalized Scene Reconstruction. In Figure 2, we show the hood and roof relit in a new environment using the reconstructed parameters. Notice how prominent the dents are under new lighting. We relight the reconstruction with a stripe projector. The relit render is tonemapped for visibility.

Figure 2. Relighting our reconstructions to emphasize that Quidient Reality can reconstruct materials as well.
Conclusions
In this blog, we demonstrated millimeter-level relightable reconstructions of automotive hail damage with Quidient Reality®. Quidient Reality® is generalized, and we plan to demonstrate this by releasing more blog posts over the coming months, featuring fine detailed relightable reconstructions of conventionally difficult scenes, among other features of our system. We have a booth at the International Conference on Computer Vision in Hawaii and would love to discuss scene reconstruction and show live reconstructions from Quidient Reality®.
Reconstruction Algorithms
Alex DeJournett
Brett Michael-Green
Brevin Tilmon
John Leffingwell
Nate Merrill
Sai Ramana Kiran Pinnama Raju
Scott Ackerson
Shailesh (Sam) Mishra



