fusion-perception-1-preview / audit /clip_detector.json
abtonmoy's picture
Add clip_detector.json to the published contamination audit
6a44e53 verified
Raw
History Blame Contribute Delete
4.59 kB
{
"detector": "openai/clip-vit-large-patch14 image tower, 224 center crop, max cosine over 140 bbox-cropped query crops",
"calibration": {
"statistic": "max cosine over all 140 bbox-cropped query descriptors",
"n_pos_images": 3439,
"n_neg_images": 11315,
"pos_mean": 0.8742,
"pos_median": 0.8932,
"neg_mean": 0.7568,
"neg_p95": 0.8888,
"tau": {
"fpr5.0": 0.8888,
"fpr1.0": 0.9145,
"fpr0.5": 0.9233,
"fpr0.1": 0.943
},
"tpr_at_tau": {
"fpr5.0": 0.5196,
"fpr1.0": 0.3955,
"fpr0.5": 0.3472,
"fpr0.1": 0.2268
}
},
"tau": 0.9145,
"tpr": 0.3955,
"fpr": 0.01,
"ours": {
"n_fetched": 2529,
"n_total": 2529,
"clip": {
"n_matched": 489,
"n_total": 2529,
"raw_precision": 0.1934,
"raw_ci95": [
0.1784,
0.2092
],
"corrected_precision": 0.4756,
"corrected_ci95": [
0.4369,
0.5168
]
},
"dinov2_same_subset": {
"n_matched": 757,
"n_total": 2529,
"raw_precision": 0.2993,
"raw_ci95": [
0.2818,
0.3175
],
"corrected_precision": 0.4255,
"corrected_ci95": [
0.3997,
0.4522
]
},
"agreement": {
"pearson_r": 0.824,
"spearman_r": 0.841,
"binary_agreement": 0.8521,
"cohens_kappa": 0.6077,
"both_positive": 436,
"clip_only": 53,
"dinov2_only": 321,
"both_negative": 1719
}
},
"rgld": {
"n_fetched": 1565,
"n_total": 1565,
"clip": {
"n_matched": 538,
"n_total": 1565,
"raw_precision": 0.3438,
"raw_ci95": [
0.3206,
0.3677
],
"corrected_precision": 0.8658,
"corrected_ci95": [
0.8058,
0.9278
]
},
"dinov2_same_subset": {
"n_matched": 870,
"n_total": 1565,
"raw_precision": 0.5559,
"raw_ci95": [
0.5312,
0.5804
],
"corrected_precision": 0.8028,
"corrected_ci95": [
0.7665,
0.8388
]
},
"agreement": {
"pearson_r": 0.6999,
"spearman_r": 0.7121,
"binary_agreement": 0.7252,
"cohens_kappa": 0.469,
"both_positive": 489,
"clip_only": 49,
"dinov2_only": 381,
"both_negative": 646
}
},
"ours_per_rule": {
"unique": {
"n_fetched": 1586,
"n_total": 1586,
"clip": {
"n_matched": 428,
"n_total": 1586,
"raw_precision": 0.2699,
"raw_ci95": [
0.2486,
0.2922
],
"corrected_precision": 0.6741,
"corrected_ci95": [
0.6189,
0.7322
]
},
"dinov2_same_subset": {
"n_matched": 644,
"n_total": 1586,
"raw_precision": 0.4061,
"raw_ci95": [
0.3821,
0.4304
],
"corrected_precision": 0.5824,
"corrected_ci95": [
0.5473,
0.6183
]
},
"agreement": {
"pearson_r": 0.8103,
"spearman_r": 0.8066,
"binary_agreement": 0.8033,
"cohens_kappa": 0.5693,
"both_positive": 380,
"clip_only": 48,
"dinov2_only": 264,
"both_negative": 894
}
},
"blanket": {
"n_fetched": 727,
"n_total": 727,
"clip": {
"n_matched": 55,
"n_total": 727,
"raw_precision": 0.0757,
"raw_ci95": [
0.0586,
0.0972
],
"corrected_precision": 0.1703,
"corrected_ci95": [
0.126,
0.2262
]
},
"dinov2_same_subset": {
"n_matched": 97,
"n_total": 727,
"raw_precision": 0.1334,
"raw_ci95": [
0.1106,
0.1601
],
"corrected_precision": 0.1815,
"corrected_ci95": [
0.148,
0.2207
]
},
"agreement": {
"pearson_r": 0.8227,
"spearman_r": 0.7813,
"binary_agreement": 0.934,
"cohens_kappa": 0.6505,
"both_positive": 52,
"clip_only": 3,
"dinov2_only": 45,
"both_negative": 627
}
},
"ctx": {
"n_fetched": 216,
"n_total": 216,
"clip": {
"n_matched": 6,
"n_total": 216,
"raw_precision": 0.0278,
"raw_ci95": [
0.0128,
0.0593
],
"corrected_precision": 0.0461,
"corrected_ci95": [
0.0072,
0.1278
]
},
"dinov2_same_subset": {
"n_matched": 16,
"n_total": 216,
"raw_precision": 0.0741,
"raw_ci95": [
0.0461,
0.1169
],
"corrected_precision": 0.0942,
"corrected_ci95": [
0.0531,
0.1572
]
},
"agreement": {
"pearson_r": 0.7212,
"spearman_r": 0.7484,
"binary_agreement": 0.9352,
"cohens_kappa": 0.3368,
"both_positive": 4,
"clip_only": 2,
"dinov2_only": 12,
"both_negative": 198
}
}
}
}