The authors constructed Achilles-600K, a region-annotated dataset of generated images with human-verified anomaly categories and bounding boxes, alongside a separate Achilles-Benchmark.
Diagnose-to-Generate aligns flow models with localized structural corrections
The authors show that detecting anomaly regions with a dedicated model before applying patch-level rewards improves structural fidelity in FLUX.1-dev and SD3.5 while preserving FID and CLIP scores.
Academic
Yifan Li · Lishuai Gao · Tong Zhang · Qilong Huang · Chen Zhao · Zhuoliang Kang · +5 more
University of Science and Technology of China · Meituan Inc.
Research Digest··2 min read
Li et al.
Why this paper
From University of Science and Technology of China and Meituan Inc.
In one line
Diagnose-to-Generate (D2G) improves structural fidelity in flow-based image generators using patch-level feedback and trajectory-preserving stochastic sampling.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks (4 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§