Feature Attribution on ImageNet 1000 samples (val) benchmark leaderboard · SOTA2 Research

Feature Attribution on ImageNet 1000 samples (val)

4.37InfidelityGuidedGradCam

Evaluation Results

Method	Links
GuidedGradCam2025.07		4.37	70.8	40.6	52.9	0.292	40.8
PullbackAscent2025.07		5.384	38.2	39.4	42.1	0.244	21.2
PullbackAscent2025.07		5.616	57	37.6	31.2	0.39	13.6
GuidedGradCam2025.07		5.761	43.7	27.5	50.7	0.242	41.7
SoftPullback2025.07		5.989	38.9	43.7	42	0.119	6.2
SmoothPullback2025.07		8.122	32.6	32.1	35.2	0.22	6.5
FusionPullback2025.07		10.131	30.6	30.6	35	0.362	5.1
SmoothPullback2025.07		14.03	54.5	42.8	30.9	0.398	4.8
FusionPullback2025.07		16.056	51.3	40.6	31.5	0.513	3.5
SoftPullback2025.07		16.303	60.6	53	37.5	0.212	6.4
GradientAscent2025.07		18.523	50.8	26.4	23.4	1.217	0
GradientAscent2025.07		29.48	27.8	13.1	23.1	1.267	0.2
FusionGrad2025.07		38.651	50.6	52.1	36.3	0.955	1.3
SmoothGrad2025.07		52.044	51.3	50.8	38.5	0.781	2.5
FusionGrad2025.07		58.514	34.4	33.6	35.7	0.855	1.4
IG2025.07		66.83	24.4	42.1	35	0.788	2.9
SmoothGrad2025.07		66.883	33.1	32.1	36.4	0.639	2.4
GradientShap2025.07		69.072	23.3	40.9	34.4	1.4	2.3
DeepLift2025.07		73.188	20.1	40.4	32.4	0.982	3.3
IG2025.07		75.796	43.4	63.4	37.9	0.752	4.6
GradientShap2025.07		76.05	43	63.2	38.1	1.086	4.2
Gradient2025.07		83.294	24.1	28.9	27.5	0.952	3.2
DeepLift2025.07		85.658	35.6	57.7	35.8	0.929	5.7
Gradient2025.07		100.09	45.7	46.6	31.6	0.898	5.3