# SAE Feature Explanation Evaluation on Vision Encoder SAE Features Synthetic Image Based Source: [SAE Feature Explanation Evaluation on Vision Encoder SAE Features Synthetic Image Based]() Collection index: [llms.txt]() Data updated: 2026-03-25T14:26:50.867Z Evaluation of the explanatory power of features generated by a Vision Encoder using Sparse Autoencoders (SAEs) on synthetic image data, assessed using metrics like IoU, AUROC, and CLIP Score. 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 | SAE Feature Explanation Evaluation | | Dataset (including any stated split) | Vision Encoder SAE Features Synthetic Image Based | | Metric | IoU (Masks) | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 6 of 6 | | 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 - [IoU (Masks)](): Higher is better; unit: PERCENT. - [IoU (Heatmaps)](): Higher is better; unit: PERCENT. - [AUROC (Masks)](): Higher is better; unit: PERCENT. - [AUROC (Heatmaps)](): Higher is better; unit: PERCENT. - [Synthetic Activation Score (Masks)](): Higher is better; unit: PERCENT. - [Synthetic Activation Score (Heatmaps)](): Higher is better; unit: PERCENT. - [CLIP Score (Masks)](): Higher is better; unit: PERCENT. - [CLIP Score (Heatmaps)](): Higher is better; unit: PERCENT. ## Recorded results | Method | IoU (Masks) | Result date | Report paper | | --- | --- | --- | --- | | Steering-informed Top-k | 22.8 | Not recorded | 69c36bfa0636e5bf4afc6406 | | Top-k | 22.4 | Not recorded | 69c36bfa0636e5bf4afc6406 | | Steering | 22 | Not recorded | 69c36bfa0636e5bf4afc6406 | | Steering-informed Top-k | 21.6 | Not recorded | 69c36bfa0636e5bf4afc6406 | | Steering | 21.1 | Not recorded | 69c36bfa0636e5bf4afc6406 | | Top-k | 21.1 | Not recorded | 69c36bfa0636e5bf4afc6406 | ### Steering-informed Top-k IoU (Masks): 22.8. Result ID: 69c3efd747f34ce924adc049. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"InternVL3"}. This record does not assert independent reproduction. ### Top-k IoU (Masks): 22.4. Result ID: 69c3efd747f34ce924adc048. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"InternVL3"}. This record does not assert independent reproduction. ### Steering IoU (Masks): 22. Result ID: 69c3efd747f34ce924adc047. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"InternVL3"}. This record does not assert independent reproduction. ### Steering-informed Top-k IoU (Masks): 21.6. Result ID: 69c3efd747f34ce924adc046. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3"}. This record does not assert independent reproduction. ### Steering IoU (Masks): 21.1. Result ID: 69c3efd747f34ce924adc044. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3"}. This record does not assert independent reproduction. ### Top-k IoU (Masks): 21.1. Result ID: 69c3efd747f34ce924adc045. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.