# Visual Question Answering on ImageNet 1k 2009 Source: [Visual Question Answering on ImageNet 1k 2009]() Collection index: [llms.txt]() Data updated: 2026-05-27T08:20:34.751Z Evaluation of Visual Question Answering performance using the ImageNet 1k 2009 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 | Visual Question Answering | | Dataset (including any stated split) | ImageNet 1k 2009 | | Metric | LLaVA | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 2 of 2 | | 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 - [LLaVA](): Higher is better; unit: PERCENT. - [BLIP-2](): Higher is better; unit: PERCENT. - [MiniGPT-4](): Higher is better; unit: PERCENT. - [Mean Performance](): Higher is better; unit: PERCENT. ## Recorded results | Method | LLaVA | Result date | Report paper | | --- | --- | --- | --- | | MMAS | 87.9 | Not recorded | 6a167aafdbd7150506f80101 | | MF-Attack | 59 | Not recorded | 6a167aafdbd7150506f80101 | ### MMAS LLaVA: 87.9. Result ID: 6a16a938fbc373592c0fbdcb. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Attack":"universal"}. This record does not assert independent reproduction. ### MF-Attack LLaVA: 59. Result ID: 6a16a938fbc373592c0fbdca. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Attack":"black-box"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.