SeeGroup: Multi-Layer Depth Estimation of Transparent Surfaces via Self-Determined Grouping

Researchers introduced SeeGroup, a new multi-layer depth estimation method for transparent surfaces.

It models depth layers as unordered events along each camera ray, using a permutation-invariant loss to learn layer groupings without fixed front-to-back assumptions.

This flexible setup improves performance over prior methods on the LayeredDepth benchmark.

The approach could help computer vision systems better separate transparent surfaces from occluded content.