Safety annotations for the ~8.7M training images used to construct the dose-response conditions (C0–C6). Two independent judges annotate the full training pool, plus a cross-classifier validation sample using all four evaluators.

No images are included. Annotations are keyed by id; pair with the original images from lehduong/flux_generated, LucasFang/FLUX-Reason-6M, and brivangl/midjourney-v6-llava.

	Repository structure… See the full description on the dataset page: https://huggingface.co/datasets/anonym371/dose-response-training-annotations.