# Faithfulness Evaluation of Image Explanation on ImageNet (val) Source: [Faithfulness Evaluation of Image Explanation on ImageNet (val)]() Collection index: [llms.txt]() Data updated: 2026-05-08T08:08:26.880Z Evaluation of the faithfulness of image explanation models on the ImageNet validation set, measured using metrics including Deletion, Insertion, and IMD. 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 | Faithfulness evaluation of image explanation | | Dataset (including any stated split) | ImageNet (val) | | Metric | Deletion | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 32 of 32 | | 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 - [Deletion](): Higher is better; unit: PERCENT. - [Insertion](): Higher is better; unit: PERCENT. - [IMD](): Higher is better; unit: SCALAR. ## Recorded results | Method | Deletion | Result date | Report paper | | --- | --- | --- | --- | | Rollout | 65.6 | Not recorded | 68fa772a1e209131cea71801 | | raw attention | 62.4 | Not recorded | 68fa772a1e209131cea71801 | | M2IB | 59.5 | Not recorded | 68fa772a1e209131cea71801 | | M2IB | 59.5 | Not recorded | 68fa772a1e209131cea71801 | | GAME | 59.4 | Not recorded | 68fa772a1e209131cea71801 | | Grad-CAM | 58.2 | Not recorded | 68fa772a1e209131cea71801 | | GAME | 57.3 | Not recorded | 68fa772a1e209131cea71801 | | Grad-CAM | 56.3 | Not recorded | 68fa772a1e209131cea71801 | | CLIPSurgery | 54.1 | Not recorded | 68fa772a1e209131cea71801 | | CLIPSurgery | 52.4 | Not recorded | 68fa772a1e209131cea71801 | | MaskCLIP | 49.6 | Not recorded | 68fa772a1e209131cea71801 | | MaskCLIP | 48.9 | Not recorded | 68fa772a1e209131cea71801 | | Grad-ECLIP w/o λi | 45.7 | Not recorded | 68fa772a1e209131cea71801 | | Grad-ECLIP | 44.2 | Not recorded | 68fa772a1e209131cea71801 | | Grad-ECLIP w/o λi | 43.8 | Not recorded | 68fa772a1e209131cea71801 | | Grad-ECLIP | 42.7 | Not recorded | 68fa772a1e209131cea71801 | | Rollout | 40.8 | Not recorded | 68fa772a1e209131cea71801 | | raw attention | 38.3 | Not recorded | 68fa772a1e209131cea71801 | | M2IB | 36.3 | Not recorded | 68fa772a1e209131cea71801 | | M2IB | 36.3 | Not recorded | 68fa772a1e209131cea71801 | | Grad-CAM | 35.2 | Not recorded | 68fa772a1e209131cea71801 | | GAME | 35 | Not recorded | 68fa772a1e209131cea71801 | | Grad-CAM | 34.2 | Not recorded | 68fa772a1e209131cea71801 | | GAME | 33.6 | Not recorded | 68fa772a1e209131cea71801 | | CLIPSurgery | 32.2 | Not recorded | 68fa772a1e209131cea71801 | | CLIPSurgery | 31.2 | Not recorded | 68fa772a1e209131cea71801 | | MaskCLIP | 28.9 | Not recorded | 68fa772a1e209131cea71801 | | MaskCLIP | 28.5 | Not recorded | 68fa772a1e209131cea71801 | | Grad-ECLIP w/o λi | 26.3 | Not recorded | 68fa772a1e209131cea71801 | | Grad-ECLIP | 25.4 | Not recorded | 68fa772a1e209131cea71801 | | Grad-ECLIP w/o λi | 25.4 | Not recorded | 68fa772a1e209131cea71801 | | Grad-ECLIP | 24.6 | Not recorded | 68fa772a1e209131cea71801 | ### Rollout Deletion: 65.6. Result ID: 69fd99f047f34ce924e077aa. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Ground-truth","Rank":"@5"}. This record does not assert independent reproduction. ### raw attention Deletion: 62.4. Result ID: 69fd99f047f34ce924e077a8. [Reporting paper]() Recorded evaluation configuration: {"Text input source":"Ground-truth","Rank":"@5"}. This record does not assert independent reproduction. ### M2IB Deletion: 59.5. Result ID: 69fd99f047f34ce924e077bc. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Text input source":"Ground-truth","Rank":"@5"}. This record does not assert independent reproduction. ### M2IB Deletion: 59.5. Result ID: 69fd99f047f34ce924e077be. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Text input source":"Prediction","Rank":"@5"}. This record does not assert independent reproduction. ### GAME Deletion: 59.4. Result ID: 69fd99f047f34ce924e077b2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Prediction","Rank":"@5"}. This record does not assert independent reproduction. ### Grad-CAM Deletion: 58.2. Result ID: 69fd99f047f34ce924e077ae. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Text input source":"Prediction","Rank":"@5"}. This record does not assert independent reproduction. ### GAME Deletion: 57.3. Result ID: 69fd99f047f34ce924e077b0. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Ground-truth","Rank":"@5"}. This record does not assert independent reproduction. ### Grad-CAM Deletion: 56.3. Result ID: 69fd99f047f34ce924e077ac. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Text input source":"Ground-truth","Rank":"@5"}. This record does not assert independent reproduction. ### CLIPSurgery Deletion: 54.1. Result ID: 69fd99f047f34ce924e077ba. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Prediction","Rank":"@5"}. This record does not assert independent reproduction. ### CLIPSurgery Deletion: 52.4. Result ID: 69fd99f047f34ce924e077b8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Ground-truth","Rank":"@5"}. This record does not assert independent reproduction. ### MaskCLIP Deletion: 49.6. Result ID: 69fd99f047f34ce924e077b6. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Prediction","Rank":"@5"}. This record does not assert independent reproduction. ### MaskCLIP Deletion: 48.9. Result ID: 69fd99f047f34ce924e077b4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Ground-truth","Rank":"@5"}. This record does not assert independent reproduction. ### Grad-ECLIP w/o λi Deletion: 45.7. Result ID: 69fd99f047f34ce924e077c6. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Prediction","Rank":"@5"}. This record does not assert independent reproduction. ### Grad-ECLIP Deletion: 44.2. Result ID: 69fd99f047f34ce924e077c2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Prediction","Rank":"@5"}. This record does not assert independent reproduction. ### Grad-ECLIP w/o λi Deletion: 43.8. Result ID: 69fd99f047f34ce924e077c4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Ground-truth","Rank":"@5"}. This record does not assert independent reproduction. ### Grad-ECLIP Deletion: 42.7. Result ID: 69fd99f047f34ce924e077c0. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Ground-truth","Rank":"@5"}. This record does not assert independent reproduction. ### Rollout Deletion: 40.8. Result ID: 69fd99f047f34ce924e077a9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Ground-truth","Rank":"@1"}. This record does not assert independent reproduction. ### raw attention Deletion: 38.3. Result ID: 69fd99f047f34ce924e077a7. [Reporting paper]() Recorded evaluation configuration: {"Text input source":"Ground-truth","Rank":"@1"}. This record does not assert independent reproduction. ### M2IB Deletion: 36.3. Result ID: 69fd99f047f34ce924e077bb. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Text input source":"Ground-truth","Rank":"@1"}. This record does not assert independent reproduction. ### M2IB Deletion: 36.3. Result ID: 69fd99f047f34ce924e077bd. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Text input source":"Prediction","Rank":"@1"}. This record does not assert independent reproduction. ### Grad-CAM Deletion: 35.2. Result ID: 69fd99f047f34ce924e077ad. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Text input source":"Prediction","Rank":"@1"}. This record does not assert independent reproduction. ### GAME Deletion: 35. Result ID: 69fd99f047f34ce924e077b1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Prediction","Rank":"@1"}. This record does not assert independent reproduction. ### Grad-CAM Deletion: 34.2. Result ID: 69fd99f047f34ce924e077ab. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Text input source":"Ground-truth","Rank":"@1"}. This record does not assert independent reproduction. ### GAME Deletion: 33.6. Result ID: 69fd99f047f34ce924e077af. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Ground-truth","Rank":"@1"}. This record does not assert independent reproduction. ### CLIPSurgery Deletion: 32.2. Result ID: 69fd99f047f34ce924e077b9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Prediction","Rank":"@1"}. This record does not assert independent reproduction. ### CLIPSurgery Deletion: 31.2. Result ID: 69fd99f047f34ce924e077b7. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Ground-truth","Rank":"@1"}. This record does not assert independent reproduction. ### MaskCLIP Deletion: 28.9. Result ID: 69fd99f047f34ce924e077b5. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Prediction","Rank":"@1"}. This record does not assert independent reproduction. ### MaskCLIP Deletion: 28.5. Result ID: 69fd99f047f34ce924e077b3. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Ground-truth","Rank":"@1"}. This record does not assert independent reproduction. ### Grad-ECLIP w/o λi Deletion: 26.3. Result ID: 69fd99f047f34ce924e077c5. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Prediction","Rank":"@1"}. This record does not assert independent reproduction. ### Grad-ECLIP Deletion: 25.4. Result ID: 69fd99f047f34ce924e077c1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Prediction","Rank":"@1"}. This record does not assert independent reproduction. ### Grad-ECLIP w/o λi Deletion: 25.4. Result ID: 69fd99f047f34ce924e077c3. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Ground-truth","Rank":"@1"}. This record does not assert independent reproduction. ### Grad-ECLIP Deletion: 24.6. Result ID: 69fd99f047f34ce924e077bf. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Text input source":"Ground-truth","Rank":"@1"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.