mirror of
https://github.com/wiltodelta/remove-ai-watermarks.git
synced 2026-10-14 21:40:47 +02:00
Victor Kuznetsov
8562c457a4
Provider classes now name the GENERATOR that rendered the pixels, never the brand. bytedance joins openai/google/muse-image as the shared ByteDance lineage (Doubao and Jimeng render with one model family; measured: separate heads cross-fire 32/22 at 3.5x mass, the surgical union holds 83.9% on the frozen test cell while openai/google/meta/no_ai stay byte-identical to run1). tc260 stops being a public value: its head remains as the group veto for the rest of China's generators (without it, 207/379 China test rows would be falsely named openai or muse-image), and the loader tolerates provider snapshots that predate bytedance. The receipt gate moves model-side: stable-name artifact receipt-gate.npz carries its threshold, head and threshold always load from one file, and the publish script stages the stable gate from the weights source. Microsoft stays metadata-only: 144 of its 253 bank rows name an OpenAI model as their C2PA softwareAgent and none names a Microsoft renderer. pre-commit: 1) maintain.sh - exit 1 (uv-secure: lightning 2.6.5 via trustmark, PYSEC-2026-3624 no fix released; per docs/development.md the core checks ran separately: ruff clean, pyright 0 errors, 2059 passed 3 skipped); 2) /simplify - single-pass four angles, clean, one noted skip (gate-test npz staging duplication); 3) docs sync - grepped bytedance/receipt-gate.npz/provider values over README, docs, skill, rules: no stale references; 4) CLAUDE.md - improver: no changes, revise: no additions (learnings live in canonical docs)
60 lines
2.7 KiB
JSON
60 lines
2.7 KiB
JSON
{
|
|
"freeze": "2026-08-31",
|
|
"task": "photo_ai_versus_camera",
|
|
"not": ["synthid_decoder", "provenance", "universal_ai_detector"],
|
|
"seeds": {
|
|
"sample": 20260821,
|
|
"detector": 20260940,
|
|
"provider": 20260956
|
|
},
|
|
"model1": {
|
|
"backbone": "openai/clip-vit-large-patch14",
|
|
"letterbox": 224,
|
|
"pad_rgb": [123, 117, 104],
|
|
"ridge_threshold_key": "thr_oi_1pct",
|
|
"ridge_threshold": 0.3056212276800537,
|
|
"mlp_threshold": 5.9586493237495395,
|
|
"mlp_width": 512,
|
|
"definitely": "ridge_and_mlp",
|
|
"public_ai_label": "definitely_only"
|
|
},
|
|
"model2": {
|
|
"feature_width": 124,
|
|
"margin": 0.3,
|
|
"classes": ["openai", "google", "bytedance", "tc260", "muse-image", "no_ai"],
|
|
"class_kind": {
|
|
"openai": "provider",
|
|
"google": "provider",
|
|
"bytedance": "provider",
|
|
"tc260": "abstain-veto",
|
|
"muse-image": "model",
|
|
"no_ai": "negative"
|
|
},
|
|
"public_values": ["openai", "google", "bytedance", "muse-image", null],
|
|
"bytedance_note": "Doubao and Jimeng render with one ByteDance model family; the class names that shared generator lineage. Measured 2026-09-07 on the frozen test cell: 83.9% (271/323), extended with the historical harvest 83.6% (529/633).",
|
|
"tc260_note": "the head covers the REST of China's generator ecosystem -- producers (Qwen, Yuanbao, Kling, ...) that are peers of openai/google/meta, not one vendor. No mixed head can honestly name the group, so its argmax win abstains (provider=None). Measured 2026-09-07: without the veto 207/379 China test rows would be falsely named openai/google/muse-image. Per-producer classes return when their train mass holds their own cells (qwen measured at 64 train rows: 27% -- stays in the fallback).",
|
|
"checkpoint_keys": {
|
|
"muse-image": "meta_muse_image"
|
|
},
|
|
"snapshot_tolerance": "provider.pt snapshots predating bytedance load with five heads; absent heads are skipped and their rows fall to the tc260 veto",
|
|
"runs_only_after": "definitely"
|
|
},
|
|
"receipt_gate": {
|
|
"added": "2026-09-02",
|
|
"asset": "receipt-gate-2026-09-02.npz",
|
|
"threshold": 2.0432573877459697,
|
|
"runs_only_after": "definitely",
|
|
"on_hit": "unknown",
|
|
"head": "linear on the CLIP-L-ft 768-d vector Model 1 already computes",
|
|
"train": "CORD-v2 train 800 (CC BY 4.0) + 200 synthetic receipts + 2400 ai_train negatives",
|
|
"certified": "min of 1st-percentile scores on CORD validation and a synthetic holdout; no field-receipt pixels in training or threshold selection",
|
|
"measured": "field receipts false-AI 19/59 -> 2/59; CORD 89/99 -> 0/99; ai_test DEFINITELY cost 7/1847"
|
|
},
|
|
"files": [
|
|
"clip-l-ft.pt",
|
|
"probe-weights-clip-l-ft.npz",
|
|
"detector.pt",
|
|
"provider.pt"
|
|
]
|
|
}
|