eval.yaml

name: PBench
description: >
  PBench is a multi-level referring expression segmentation benchmark that
  evaluates vision-language perception across a structured hierarchy of skills.
  It contains 6,338 samples with 83,977 instance masks across 5,090 unique
  expressions, organized into five complexity levels plus a dense split.
  Metric: macro F1.
evaluation_framework: pbench

tasks:
  - id: level_0
    config: default
    split: level_0
  - id: level_1
    config: default
    split: level_1
  - id: level_2
    config: default
    split: level_2
  - id: level_3
    config: default
    split: level_3
  - id: level_4
    config: default
    split: level_4
  - id: dense
    config: default
    split: dense
  - id: average
    config: default
    split: level_0