Calibration on CIFAR-10 5000-sample half (test) benchmark leaderboard · SOTA2 Research

Calibration on CIFAR-10 5000-sample half (test)

0.0095ECEBBQ

Evaluation Results

Method	Links
BBQ2026.05		0.0095	0.0147	0.431	0.197	0.0185
RCMMC2026.05		0.01	0.0093	0.397	0.192	0.0205
AR-CondCal2026.05		0.0105	0.0103	0.407	0.194	0.0193
Vector Scaling2026.05		0.0109	0.0119	0.391	0.19	0.0186
Classwise TS2026.05		0.0109	0.0099	0.397	0.192	0.0191
Conf + PredEntropy2026.05		0.0115	0.0105	0.406	0.194	0.0207
Conf-only2026.05		0.0117	0.0114	0.402	0.193	0.0201
Histogram Binning2026.05		0.0122	0.0144	0.428	0.197	0.0206
Classwise TS2026.05		0.0123	0.0151	0.494	0.232	0.0177
Temp. Scaling (TS)2026.05		0.0124	0.0132	0.397	0.192	0.0179
Ensemble TS2026.05		0.0124	0.0132	0.397	0.192	0.0179
Parametric TS2026.05		0.0126	0.0116	0.4	0.193	0.0171
Parametric TS2026.05		0.013	0.0125	0.498	0.234	0.024
SB-ECE TS2026.05		0.0135	0.0137	0.399	0.192	0.0198
Isotonic Regression2026.05		0.0137	0.0112	0.437	0.197	0.0209
Histogram Binning2026.05		0.0138	0.0143	0.531	0.238	0.021
Vector Scaling2026.05		0.0145	0.0151	0.483	0.229	0.0176
BBQ2026.05		0.0148	0.0219	0.539	0.239	0.0216
Temp. Scaling (TS)2026.05		0.0151	0.0175	0.496	0.233	0.0251
Ensemble TS2026.05		0.0151	0.0175	0.496	0.233	0.0251
Isotonic Regression2026.05		0.0152	0.0158	0.538	0.238	0.0206
No calibration2026.05		0.1423	0.1422	0.515	0.219	0.1536
No calibration2026.05		0.1729	0.1728	0.64	0.269	0.1783