# Visual Question Answering (Edit Inspection) on EditInspector 1.0 (test) Source: [Visual Question Answering (Edit Inspection) on EditInspector 1.0 (test)]() Collection index: [llms.txt]() Data updated: 2026-02-26T05:27:00.169Z Evaluation of model performance on the EditInspector 1.0 dataset (test split) for assessing image edits using VQA focused on inspection criteria. 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 | Visual Question Answering (Edit Inspection) | | Dataset (including any stated split) | EditInspector 1.0 (test) | | Metric | Accuracy | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 9 of 9 | | 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 - [Accuracy](): Higher is better; unit: PERCENT. - [Contextual Consistency](): Higher is better; unit: PERCENT. - [Technical Precision](): Higher is better; unit: PERCENT. - [Artifacts Score](): Higher is better; unit: PERCENT. - [Diff Caption Acc](): Higher is better; unit: PERCENT. ## Recorded results | Method | Accuracy | Result date | Report paper | | --- | --- | --- | --- | | Gemini-1.5 | 70.3 | Not recorded | 68fa79651e209131cea7d6a6 | | InternVL2 | 70.2 | Not recorded | 68fa79651e209131cea7d6a6 | | GPT-4o | 67.8 | Not recorded | 68fa79651e209131cea7d6a6 | | Qwen2.5-VL | 67.7 | Not recorded | 68fa79651e209131cea7d6a6 | | GPT-4 | 67.3 | Not recorded | 68fa79651e209131cea7d6a6 | | LLaVA | 67.2 | Not recorded | 68fa79651e209131cea7d6a6 | | GPT-4 Turbo | 66.9 | Not recorded | 68fa79651e209131cea7d6a6 | | LLaVA | 58.9 | Not recorded | 68fa79651e209131cea7d6a6 | | Gemini | 49.9 | Not recorded | 68fa79651e209131cea7d6a6 | ### Gemini-1.5 Accuracy: 70.3. Result ID: 698b4b8e8d67a41d0a6ce170. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"mode":"zero-shot"}. This record does not assert independent reproduction. ### InternVL2 Accuracy: 70.2. Result ID: 698b4b8e8d67a41d0a6ce175. [Reporting paper]() Recorded evaluation configuration: {"mode":"zero-shot"}. This record does not assert independent reproduction. ### GPT-4o Accuracy: 67.8. Result ID: 698b4b8e8d67a41d0a6ce172. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"mode":"zero-shot"}. This record does not assert independent reproduction. ### Qwen2.5-VL Accuracy: 67.7. Result ID: 698b4b8e8d67a41d0a6ce174. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"mode":"zero-shot"}. This record does not assert independent reproduction. ### GPT-4 Accuracy: 67.3. Result ID: 698b4b8e8d67a41d0a6ce171. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"mode":"zero-shot"}. This record does not assert independent reproduction. ### LLaVA Accuracy: 67.2. Result ID: 698b4b8e8d67a41d0a6ce177. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Supervised":true}. This record does not assert independent reproduction. ### GPT-4 Turbo Accuracy: 66.9. Result ID: 698b4b8e8d67a41d0a6ce173. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"mode":"zero-shot"}. This record does not assert independent reproduction. ### LLaVA Accuracy: 58.9. Result ID: 698b4b8e8d67a41d0a6ce176. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"mode":"zero-shot"}. This record does not assert independent reproduction. ### Gemini Accuracy: 49.9. Result ID: 698b4b8e8d67a41d0a6ce16f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"mode":"zero-shot"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.