Add DepthAgent

[jiezhu23]

Summary

• Add DepthAgent to Section 3.1, Information Seeking and Media Understanding.
• Follow the end-to-end system template and chronological ordering.
• Link the primary arXiv record.

Canonical placement

DepthAgent is an end-to-end multimodal agent rather than only a training method. It reasons over image, scene, and camera cues, invokes frozen depth experts as tools, and observes returned depth maps and auxiliary depth features. Those observations causally affect the next action: invoking another expert, comparing complementary predictions, or terminating with a selected or fused result.

Checks

• Confirmed no duplicate by title or arXiv ID.
• Confirmed the newly added arXiv link returns HTTP 200.
• awesome-lint reaches the repository-wide pre-existing license-detection error only.
• The current markdown-link-check release cannot start under the local Node 20 runtime because a dependency now requires Node 22; no environment was changed.

Disclosure

I am an author of DepthAgent.