The Value of Quality

Quality In

During a brief visit to London’s Victoria and Albert Museum one of the subjects was shot with the best camera to hand – an iPhone.

The museum contains a wealth of art – the very finest expressions of what it means to be a creator. There were many worthy subjects but most were behind glass or in dimly lit rooms. However, the Casts Collection was accessible and had fairly even lighting.

Many objects – such as the cast of Micaelangelo’s David – were not going to be scanned in the short time available but one object took under 100 frames and just 6 minutes. Here’s the result:

Tomb of Ilaria del Carretto – V&A Museum

Right First Time?

The model above is the second go at getting it right. The first model threw up a quality issue. The surface around the subject’s left thigh looked as if the plaster had been attacked by acid.

Something has gone wrong with the mesh – the original does not look like this.

Something has gone wrong, but what is the cause? The last blog post about talks about quality of data in = good data out so the first place to start is looking at the images. Three images were all aligned at this point.

Note the multiple images stacked on top of each other – what is going on here?
Three images, all aligned and taken in very close proximity to each other.

It was time to have a look at the images individually. The first image was clearly blurred – camera shake – and this was going to cause depth map issues.

Zoomed and cropped from the original to highlight the issue – this image is blurred thanks to camera shake.

The image aligned but is doing harm to the depth map and mesh. The next image was better, but still blurred:

This image is better but still not sharp enough.

Better, but still not good enough. This image will need disabling. The third image proved to be the winner:

This one is pin-sharp and the only one needed.

More Than One Frame?

Three images exist, all taken at roughly the same point. Knowing the first two might turn out blurred is just down to experience, so multiple shots were taken to make sure ‘the one’ was in the bag. Realising there might be duff images in the batch was forgotten and the simple action of checking for image quality before starting the 3D model build was omitted. The results were for all to see – a mess. Metashape can be very forgiving with image quality when aligning, but this can store up issues later, particularly with depth map mesh generation. Out of curiosity a dense cloud version was run and visually the model appeared better, but still not perfect.

The two offending cameras were disabled and the mesh regenerated. The section version turned out as expected with crisp detail and the end result a faithful rendering of the object in the collection.

So, before rushing into building the model spend a few minutes verifying what is shot is suitable. It will save time in the long run. In the meantime, here’s a link to the initial version: