Evaluation inputs and manifests for
Phoenix, accompanying the ECCV 2026
paper Learning from Adversity: Semantic-Aware Mask Refinement through
Adversarial Perturbation.
This repository is not a standalone copy of the benchmark datasets. It is
designed to be combined locally with original datasets downloaded by each user
from their official sources.
