Upload reliability_refresh_results.json with huggingface_hub
Browse files- reliability_refresh_results.json +115 -0
reliability_refresh_results.json
ADDED
|
@@ -0,0 +1,115 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"methodology_note": "RQ1-only refresh (ECE + reliability diagram) against whatever checkpoint currently sits at best_{model}.pth -- i.e. the promoted dropout-fix checkpoints for xception/vit, deployed 2026-07-25. Same held-out val split (seed=42, val_fraction=0.15) as every other analysis in this project. RQ2/RQ3 unaffected by the dropout-fix and intentionally not re-run here.",
|
| 3 |
+
"val_frame_count": 11666,
|
| 4 |
+
"val_video_count": 1049,
|
| 5 |
+
"models": {
|
| 6 |
+
"xception": {
|
| 7 |
+
"frame_ece": 0.07310491284879563,
|
| 8 |
+
"video_ece": 0.0832806306958424,
|
| 9 |
+
"frame_auroc": 0.9926626311220669,
|
| 10 |
+
"video_auroc": 0.9983982202447164,
|
| 11 |
+
"frame_reliability": {
|
| 12 |
+
"bin_accuracies": [
|
| 13 |
+
0.8854988001371272,
|
| 14 |
+
0.2261904761904762,
|
| 15 |
+
0.14960629921259844,
|
| 16 |
+
0.208,
|
| 17 |
+
0.05128205128205128,
|
| 18 |
+
0.9420289855072463,
|
| 19 |
+
0.9622641509433962,
|
| 20 |
+
1.0,
|
| 21 |
+
0.9887323943661972,
|
| 22 |
+
1.0
|
| 23 |
+
],
|
| 24 |
+
"bin_confidences": [
|
| 25 |
+
0.9913699195327876,
|
| 26 |
+
0.8552316016118441,
|
| 27 |
+
0.7487110670391969,
|
| 28 |
+
0.649741117477417,
|
| 29 |
+
0.5505873545622214,
|
| 30 |
+
0.548205548438473,
|
| 31 |
+
0.6523496576075284,
|
| 32 |
+
0.755149797157005,
|
| 33 |
+
0.855771786562154,
|
| 34 |
+
0.9926880709155389
|
| 35 |
+
],
|
| 36 |
+
"bin_counts": [
|
| 37 |
+
2917,
|
| 38 |
+
168,
|
| 39 |
+
127,
|
| 40 |
+
125,
|
| 41 |
+
117,
|
| 42 |
+
138,
|
| 43 |
+
159,
|
| 44 |
+
189,
|
| 45 |
+
355,
|
| 46 |
+
7045
|
| 47 |
+
],
|
| 48 |
+
"bin_boundaries": [
|
| 49 |
+
0.0,
|
| 50 |
+
0.1,
|
| 51 |
+
0.2,
|
| 52 |
+
0.30000000000000004,
|
| 53 |
+
0.4,
|
| 54 |
+
0.5,
|
| 55 |
+
0.6000000000000001,
|
| 56 |
+
0.7000000000000001,
|
| 57 |
+
0.8,
|
| 58 |
+
0.9,
|
| 59 |
+
1.0
|
| 60 |
+
]
|
| 61 |
+
},
|
| 62 |
+
"video_reliability": {
|
| 63 |
+
"bin_accuracies": [
|
| 64 |
+
0.96,
|
| 65 |
+
0.3333333333333333,
|
| 66 |
+
0.0,
|
| 67 |
+
0.125,
|
| 68 |
+
0.0,
|
| 69 |
+
1.0,
|
| 70 |
+
1.0,
|
| 71 |
+
1.0,
|
| 72 |
+
1.0,
|
| 73 |
+
1.0
|
| 74 |
+
],
|
| 75 |
+
"bin_confidences": [
|
| 76 |
+
0.9878810474176987,
|
| 77 |
+
0.8538837263479946,
|
| 78 |
+
0.7486330546078864,
|
| 79 |
+
0.6525810244027526,
|
| 80 |
+
0.5475465111155062,
|
| 81 |
+
0.5623856821957894,
|
| 82 |
+
0.6473097034453295,
|
| 83 |
+
0.7552529573418257,
|
| 84 |
+
0.8575347691066373,
|
| 85 |
+
0.9847406328305219
|
| 86 |
+
],
|
| 87 |
+
"bin_counts": [
|
| 88 |
+
150,
|
| 89 |
+
15,
|
| 90 |
+
11,
|
| 91 |
+
8,
|
| 92 |
+
10,
|
| 93 |
+
23,
|
| 94 |
+
34,
|
| 95 |
+
47,
|
| 96 |
+
98,
|
| 97 |
+
650
|
| 98 |
+
],
|
| 99 |
+
"bin_boundaries": [
|
| 100 |
+
0.0,
|
| 101 |
+
0.1,
|
| 102 |
+
0.2,
|
| 103 |
+
0.30000000000000004,
|
| 104 |
+
0.4,
|
| 105 |
+
0.5,
|
| 106 |
+
0.6000000000000001,
|
| 107 |
+
0.7000000000000001,
|
| 108 |
+
0.8,
|
| 109 |
+
0.9,
|
| 110 |
+
1.0
|
| 111 |
+
]
|
| 112 |
+
}
|
| 113 |
+
}
|
| 114 |
+
}
|
| 115 |
+
}
|