Out Of Distribution Detection On Imagenet 1K 9

评估指标

AUROC

FPR95

评测结果

各个模型在此基准测试上的表现结果

			Paper Title	Repository
RP+GradNorm	-	83.29	Detecting Out-of-distribution Data through In-distribution Class Prior	-
KNN (ResNet-50)	74.87	77.09	Out-of-Distribution Detection with Deep Nearest Neighbors
Watermarking (WRN-40-2 w/ Energy)	79.85	71.85	Watermarking for Out-of-distribution Detection
Watermarking (WRN-40-2 w/ MSP)	82.03	70.59	Watermarking for Out-of-distribution Detection
DOE	83.05	67.84	Out-of-distribution Detection with Implicit Outlier Transformation
DML	-	61.43	Decoupling MaxLogit for Out-of-Distribution Detection	-
GradNorm (ResNetv2-101)	-	60.86	On the Importance of Gradients for Detecting Distributional Shifts in the Wild
KNN (ResNet-50 SupCon)	84.62	60.02	Out-of-Distribution Detection with Deep Nearest Neighbors
ReAct (ResNet-50)	86.64	51.56	ReAct: Out-of-distribution Detection With Rectified Activations
ODIN+UMAP (ResNet-50)	86.99	50.06	Unleashing Mask: Explore the Intrinsic Out-of-Distribution Detection Capability
MOS (BiT-S-R101x1)	89.06	49.54	MOS: Towards Scaling Out-of-distribution Detection for Large Semantic Space
DICE (ResNet-50)	87.48	46.49	DICE: Leveraging Sparsification for Out-of-Distribution Detection
SHE	-	45.35	Out-of-Distribution Detection based on In-Distribution Data Patterns Memorization with Modern Hopfield Energy	-
NPOS	89.44	45.27	Non-Parametric Outlier Synthesis
MCM (CLIP-B)	89.77	44.69	Delving into Out-of-Distribution Detection with Vision-Language Representations
ASH-S (ResNet-50)	90.98	39.78	Extremely Simple Activation Shaping for Out-of-Distribution Detection
RankFeat (ResNetv2-101)	90.93	39.34	RankFeat: Rank-1 Feature Removal for Out-of-distribution Detection
DICE + ReAct (ResNet-50)	90.67	36.86	DICE: Leveraging Sparsification for Out-of-Distribution Detection
MCM (CLIP-L)	92.00	35.42	Delving into Out-of-Distribution Detection with Vision-Language Representations
SCALE (ResNet50)	92.26	34.51	Scaling for Training Time and Post-hoc Out-of-distribution Detection Enhancement

0 of 24 row(s) selected.