Benchmark RadarAI BENCHMARK PROFILE
SOCO
SOCO is a benchmark for semantic object correspondence with keypoint annotations across 100 categories and over 1M pairs. The evaluation is implemented within OmniProbe, which includes correspondence tasks among others.
- Released
- 2026-05-29
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
It provides a systematic evaluation for part-level understanding in vision models, but since it is part of a larger framework, it is not a standalone benchmark.
Motivation
Measuring structured object understanding in vision foundation models remains challenging due to inconsistent evaluation protocols and limited part-level supervision.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.