# VQA Hallucination on POPE MS COCO v1 (test) Source: [VQA Hallucination on POPE MS COCO v1 (test)]() Collection index: [llms.txt]() Data updated: 2026-02-26T06:00:56.694Z Evaluation of model VQA hallucination performance using the POPE methodology on the MS COCO v1 test set, reporting accuracy across Random, Popular, and Adversarial query types. 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 | VQA Hallucination | | Dataset (including any stated split) | POPE MS COCO v1 (test) | | Metric | Random Accuracy | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 11 of 11 | | 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 - [Random Accuracy](): Higher is better; unit: PERCENT. - [Random Yes Accuracy](): Higher is better; unit: PERCENT. - [Popular Accuracy](): Higher is better; unit: PERCENT. - [Popular Yes Accuracy](): Higher is better; unit: PERCENT. - [Adversarial Accuracy](): Higher is better; unit: PERCENT. - [Adversarial Yes Accuracy](): Higher is better; unit: PERCENT. - [Overall Accuracy](): Higher is better; unit: PERCENT. - [Overall Yes Accuracy](): Higher is better; unit: PERCENT. ## Recorded results | Method | Random Accuracy | Result date | Report paper | | --- | --- | --- | --- | | Robust mPLUG-Owl-7B | 86 | Not recorded | 68fa729b1e209131cea5779f | | LLaVA-RLHF\_13B | 85.2 | Not recorded | 68fa729b1e209131cea5779f | | LLaVA\_13B M3ID + DPO | 85.2 | Not recorded | 68fa729b1e209131cea5779f | | LLaVA-RLHF\_7B | 84.8 | Not recorded | 68fa729b1e209131cea5779f | | LLaVA\_13B M3ID | 84.3 | Not recorded | 68fa729b1e209131cea5779f | | LLaVA\_7B M3ID + DPO | 81.2 | Not recorded | 68fa729b1e209131cea5779f | | LLaVA\_7B M3ID | 76 | Not recorded | 68fa729b1e209131cea5779f | | LLaVA\_7B | 74.8 | Not recorded | 68fa729b1e209131cea5779f | | MiniGPT4-13B | 73 | Not recorded | 68fa729b1e209131cea5779f | | LLaVA\_13B | 67.9 | Not recorded | 68fa729b1e209131cea5779f | | mPLUG-Owl-7B | 52 | Not recorded | 68fa729b1e209131cea5779f | ### Robust mPLUG-Owl-7B Random Accuracy: 86. Result ID: 699262b03a6c3148490e9526. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"7B","Supervised Training (SUP)":true}. This record does not assert independent reproduction. ### LLaVA-RLHF\_13B Random Accuracy: 85.2. Result ID: 699262b03a6c3148490e952c. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"13B","Supervised Training (SUP)":true}. This record does not assert independent reproduction. ### LLaVA\_13B M3ID + DPO Random Accuracy: 85.2. Result ID: 699262b03a6c3148490e9530. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"13B","Decoding Strategy":"M3ID","Training":"DPO"}. This record does not assert independent reproduction. ### LLaVA-RLHF\_7B Random Accuracy: 84.8. Result ID: 699262b03a6c3148490e9527. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"7B","Supervised Training (SUP)":true}. This record does not assert independent reproduction. ### LLaVA\_13B M3ID Random Accuracy: 84.3. Result ID: 699262b03a6c3148490e952f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"13B","Decoding Strategy":"M3ID (Inference-time intervention)"}. This record does not assert independent reproduction. ### LLaVA\_7B M3ID + DPO Random Accuracy: 81.2. Result ID: 699262b03a6c3148490e952b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"7B","Decoding Strategy":"M3ID","Training":"DPO"}. This record does not assert independent reproduction. ### LLaVA\_7B M3ID Random Accuracy: 76. Result ID: 699262b03a6c3148490e952a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"7B","Decoding Strategy":"M3ID (Inference-time intervention)"}. This record does not assert independent reproduction. ### LLaVA\_7B Random Accuracy: 74.8. Result ID: 699262b03a6c3148490e9529. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"7B","Decoding Strategy":"Standard"}. This record does not assert independent reproduction. ### MiniGPT4-13B Random Accuracy: 73. Result ID: 699262b03a6c3148490e952d. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"13B"}. This record does not assert independent reproduction. ### LLaVA\_13B Random Accuracy: 67.9. Result ID: 699262b03a6c3148490e952e. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"13B","Decoding Strategy":"Standard"}. This record does not assert independent reproduction. ### mPLUG-Owl-7B Random Accuracy: 52. Result ID: 699262b03a6c3148490e9528. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"7B","Supervised Training (SUP)":false}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.