# XAI Evaluation on ImageNet-S (val) Source: [XAI Evaluation on ImageNet-S (val)]() Collection index: [llms.txt]() Data updated: 2026-02-27T11:24:26.093Z Evaluation of Explainable AI (XAI) methods, including attribution fidelity and localization metrics, performed specifically on the validation split of the ImageNet-S dataset. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | XAI Evaluation | | Dataset (including any stated split) | ImageNet-S (val) | | Metric | Selection Score | | Unit | SCALAR | | Direction | Higher is better | | Results loaded | 28 of 28 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [Selection Score](): Higher is better; unit: SCALAR. - [Attribution Localization](): Higher is better; unit: SCALAR. - [Point Score](): Higher is better; unit: SCALAR. - [Pixel Flips AUC@5](): Higher is better; unit: PERCENT. - [Pixel Flips AUC@20](): Higher is better; unit: SCALAR. - [Max Sensitivity](): Higher is better; unit: SCALAR. ## Recorded results | Method | Selection Score | Result date | Report paper | | --- | --- | --- | --- | | Feature Ablation | 6.029 | Not recorded | 6965d1a46343fa78c8323821 | | GradCAM | 6.006 | Not recorded | 6965d1a46343fa78c8323821 | | GradCAM++ | 5.955 | Not recorded | 6965d1a46343fa78c8323821 | | SplitGrad (α 0.4) | 5.633 | Not recorded | 6965d1a46343fa78c8323821 | | Deconvolution | 5.435 | Not recorded | 6965d1a46343fa78c8323821 | | Occlusion | 5.214 | Not recorded | 6965d1a46343fa78c8323821 | | SplitLRP (sc, wta) | 5.168 | Not recorded | 6965d1a46343fa78c8323821 | | Guided GradCAM | 5.13 | Not recorded | 6965d1a46343fa78c8323821 | | SplitGrad (α 0.5, wta) | 4.948 | Not recorded | 6965d1a46343fa78c8323821 | | Integrated Gradients | 4.838 | Not recorded | 6965d1a46343fa78c8323821 | | Gradient SHAP | 4.808 | Not recorded | 6965d1a46343fa78c8323821 | | SplitGrad (α 0.4, wta) | 4.735 | Not recorded | 6965d1a46343fa78c8323821 | | SplitCAM (sc, α 0.4) | 4.72 | Not recorded | 6965d1a46343fa78c8323821 | | SplitCAM (sc, α 0.4, wta) | 4.711 | Not recorded | 6965d1a46343fa78c8323821 | | SplitCAM (sc, α 0.5, wta) | 4.608 | Not recorded | 6965d1a46343fa78c8323821 | | DeepLift | 4.447 | Not recorded | 6965d1a46343fa78c8323821 | | SplitLRP (sc) | 4.393 | Not recorded | 6965d1a46343fa78c8323821 | | Guided GradCAM | 4.301 | Not recorded | 6965d1a46343fa78c8323821 | | Guided Backprop | 3.448 | Not recorded | 6965d1a46343fa78c8323821 | | SplitGrad (α 0.4) | 3.429 | Not recorded | 6965d1a46343fa78c8323821 | | SplitGrad (α 0.5, wta) | 3.379 | Not recorded | 6965d1a46343fa78c8323821 | | SplitGrad (α 0.4, wta) | 3.287 | Not recorded | 6965d1a46343fa78c8323821 | | Guided Backprop | 3.276 | Not recorded | 6965d1a46343fa78c8323821 | | SplitLRP (sc, wta) | 3.235 | Not recorded | 6965d1a46343fa78c8323821 | | SplitLRP (sc) | 3.062 | Not recorded | 6965d1a46343fa78c8323821 | | SplitCAM (sc, α 0.5, wta) | 2.766 | Not recorded | 6965d1a46343fa78c8323821 | | SplitCAM (sc, α 0.4, wta) | 2.761 | Not recorded | 6965d1a46343fa78c8323821 | | SplitCAM (sc, α 0.4) | 2.727 | Not recorded | 6965d1a46343fa78c8323821 | ### Feature Ablation Selection Score: 6.029. Result ID: 69a17edd35dfcabb00e76e6e. [Reporting paper]() Recorded evaluation configuration: {"Layer":"0","Abs":"y","Backbone":"VGG16"}. This record does not assert independent reproduction. ### GradCAM Selection Score: 6.006. Result ID: 69a17edd35dfcabb00e76e77. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Layer":"0","Abs":"n","Backbone":"VGG16"}. This record does not assert independent reproduction. ### GradCAM++ Selection Score: 5.955. Result ID: 69a17edd35dfcabb00e76e72. [Reporting paper]() Recorded evaluation configuration: {"Layer":"30","Abs":"n","Backbone":"VGG16"}. This record does not assert independent reproduction. ### SplitGrad (α 0.4) Selection Score: 5.633. Result ID: 69a17edd35dfcabb00e76e62. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"(12, +, g)","Abs":"n","Backbone":"VGG16"}. This record does not assert independent reproduction. ### Deconvolution Selection Score: 5.435. Result ID: 69a17edd35dfcabb00e76e6c. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"0","Abs":"y","Backbone":"VGG16"}. This record does not assert independent reproduction. ### Occlusion Selection Score: 5.214. Result ID: 69a17edd35dfcabb00e76e71. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"0","Abs":"y","Backbone":"VGG16"}. This record does not assert independent reproduction. ### SplitLRP (sc, wta) Selection Score: 5.168. Result ID: 69a17edd35dfcabb00e76e6b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"(26, pos)","Abs":"y","Backbone":"VGG16"}. This record does not assert independent reproduction. ### Guided GradCAM Selection Score: 5.13. Result ID: 69a17edd35dfcabb00e76e76. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Layer":"26","Abs":"n","Backbone":"VGG16"}. This record does not assert independent reproduction. ### SplitGrad (α 0.5, wta) Selection Score: 4.948. Result ID: 69a17edd35dfcabb00e76e67. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"(26, +, g)","Abs":"y","Backbone":"VGG16"}. This record does not assert independent reproduction. ### Integrated Gradients Selection Score: 4.838. Result ID: 69a17edd35dfcabb00e76e70. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"0","Abs":"y","Backbone":"VGG16"}. This record does not assert independent reproduction. ### Gradient SHAP Selection Score: 4.808. Result ID: 69a17edd35dfcabb00e76e6f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"0","Abs":"y","Backbone":"VGG16"}. This record does not assert independent reproduction. ### SplitGrad (α 0.4, wta) Selection Score: 4.735. Result ID: 69a17edd35dfcabb00e76e64. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"(12, +, g)","Abs":"n","Backbone":"VGG16"}. This record does not assert independent reproduction. ### SplitCAM (sc, α 0.4) Selection Score: 4.72. Result ID: 69a17edd35dfcabb00e76e5c. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"(12, g)","Abs":"n","Backbone":"VGG16"}. This record does not assert independent reproduction. ### SplitCAM (sc, α 0.4, wta) Selection Score: 4.711. Result ID: 69a17edd35dfcabb00e76e5f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"(26, +, g)","Abs":"y","Backbone":"VGG16"}. This record does not assert independent reproduction. ### SplitCAM (sc, α 0.5, wta) Selection Score: 4.608. Result ID: 69a17edd35dfcabb00e76e61. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"(26, +, g)","Abs":"n","Backbone":"VGG16"}. This record does not assert independent reproduction. ### DeepLift Selection Score: 4.447. Result ID: 69a17edd35dfcabb00e76e6d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"0","Abs":"y","Backbone":"VGG16"}. This record does not assert independent reproduction. ### SplitLRP (sc) Selection Score: 4.393. Result ID: 69a17edd35dfcabb00e76e69. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"(19, pos)","Abs":"n","Backbone":"VGG16"}. This record does not assert independent reproduction. ### Guided GradCAM Selection Score: 4.301. Result ID: 69a17edd35dfcabb00e76e75. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Layer":"0","Abs":"n","Backbone":"VGG16"}. This record does not assert independent reproduction. ### Guided Backprop Selection Score: 3.448. Result ID: 69a17edd35dfcabb00e76e74. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"5","Abs":"n","Backbone":"VGG16"}. This record does not assert independent reproduction. ### SplitGrad (α 0.4) Selection Score: 3.429. Result ID: 69a17edd35dfcabb00e76e63. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"(14, +, g)","Abs":"y","Backbone":"VGG16"}. This record does not assert independent reproduction. ### SplitGrad (α 0.5, wta) Selection Score: 3.379. Result ID: 69a17edd35dfcabb00e76e66. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"(12, +, g)","Abs":"y","Backbone":"VGG16"}. This record does not assert independent reproduction. ### SplitGrad (α 0.4, wta) Selection Score: 3.287. Result ID: 69a17edd35dfcabb00e76e65. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"(14, +, g)","Abs":"y","Backbone":"VGG16"}. This record does not assert independent reproduction. ### Guided Backprop Selection Score: 3.276. Result ID: 69a17edd35dfcabb00e76e73. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"2","Abs":"n","Backbone":"VGG16"}. This record does not assert independent reproduction. ### SplitLRP (sc, wta) Selection Score: 3.235. Result ID: 69a17edd35dfcabb00e76e6a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"(7, pos)","Abs":"n","Backbone":"VGG16"}. This record does not assert independent reproduction. ### SplitLRP (sc) Selection Score: 3.062. Result ID: 69a17edd35dfcabb00e76e68. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"(7, pos)","Abs":"n","Backbone":"VGG16"}. This record does not assert independent reproduction. ### SplitCAM (sc, α 0.5, wta) Selection Score: 2.766. Result ID: 69a17edd35dfcabb00e76e60. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"(14, g)","Abs":"n","Backbone":"VGG16"}. This record does not assert independent reproduction. ### SplitCAM (sc, α 0.4, wta) Selection Score: 2.761. Result ID: 69a17edd35dfcabb00e76e5e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"(14, g)","Abs":"n","Backbone":"VGG16"}. This record does not assert independent reproduction. ### SplitCAM (sc, α 0.4) Selection Score: 2.727. Result ID: 69a17edd35dfcabb00e76e5d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Layer":"(14, g)","Abs":"n","Backbone":"VGG16"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.