Campaign 2

Description

Individual type 30B cylinders:

_images/8342.png _images/21967.png _images/49792.png

Statistics

Total samples: 54999.

  • Samples with provenance: 0

  • Samples with images: 54999

  • Samples with synthetic data: 54999

  • Samples with cryptomatte data: 49999

  • Samples with bounding-boxes: 49067

  • Samples with contours: 49067

  • Samples with tags: 54999

Categories:

  • Category 30B samples: 49067

  • Category background samples: 5000

  • Category synthetic samples: 54999

Annotations:

  • Samples annotated synthetic: 932

  • Samples annotated background + synthetic: 5000

  • Samples annotated 30B + synthetic: 49067

Category Definitions

  • 30B: type 30B cylinders.

  • background: samples that intentionally contain only background, without any cylinders.

  • synthetic: samples generated using our synthetic workflow.

Notes

There is a fairly wide range of distances from camera to cylinder, so that they do not tend to fill the frame.

Backgrounds and non-backgrounds are aligned: for all \(n\) where \(n \bmod 10 = 0\), each sample image_\(n\) with a cylinder has a matching sample background_\(n\) without a cylinder.

Questions

What happened to Campaign 1?

Campaign 1 was for development and debugging of our synthetic workflow.

Why are there 54999 samples, instead of 55000?

We discovered that one output sample had garbled cryptomatte data, so we deleted it.

Why are there 932 samples without a container that are tagged “synthetic” but not “background”?

Those are scenes which contained a type 30B cylinder which would otherwise be tagged “30B + synthetic”, except that randomly-chosen camera angle put the cylinder off-screen. So the labelling is accurate.