# Video Question Answering on VideoEspresso 1.0 Source: [Video Question Answering on VideoEspresso 1.0]() Collection index: [llms.txt]() Data updated: 2026-02-26T05:53:32.819Z Evaluation of model performance on the VideoEspresso 1.0 dataset for Video Question Answering, reporting performance across various sub-categories (e.g., Narrative, Causal, Thematic) and an overall average. 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 | Video Question Answering | | Dataset (including any stated split) | VideoEspresso 1.0 | | Metric | Narrative Accuracy | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 13 of 13 | | 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 - [Narrative Accuracy](): Higher is better; unit: PERCENT. - [Event Accuracy](): Higher is better; unit: PERCENT. - [Ingredient Accuracy](): Higher is better; unit: PERCENT. - [Causal Accuracy](): Higher is better; unit: PERCENT. - [Thematic Accuracy](): Higher is better; unit: PERCENT. - [Content Accuracy](): Higher is better; unit: PERCENT. - [Influence Accuracy](): Higher is better; unit: PERCENT. - [Role Accuracy](): Higher is better; unit: PERCENT. - [Interaction Accuracy](): Higher is better; unit: PERCENT. - [Behavior Accuracy](): Higher is better; unit: PERCENT. - [Emotion Accuracy](): Higher is better; unit: PERCENT. - [Cooking Accuracy](): Higher is better; unit: PERCENT. - [Traffic Accuracy](): Higher is better; unit: PERCENT. - [Situation Accuracy](): Higher is better; unit: PERCENT. - [Average Accuracy](): Higher is better; unit: PERCENT. ## Recorded results | Method | Narrative Accuracy | Result date | Report paper | | --- | --- | --- | --- | | GPT-4o + ReViSe | 51.9 | Not recorded | 6993f5e79ebc499c974098f3 | | VideoEspresso | 45.2 | Not recorded | 6993f5e79ebc499c974098f3 | | Qwen2-VL + ReViSe | 39.2 | Not recorded | 6993f5e79ebc499c974098f3 | | LongVA-DPO | 35.5 | Not recorded | 6993f5e79ebc499c974098f3 | | InternVL2 + ReViSe | 35.5 | Not recorded | 6993f5e79ebc499c974098f3 | | InternVL2 | 33.9 | Not recorded | 6993f5e79ebc499c974098f3 | | Qwen-VL-Max | 33.9 | Not recorded | 6993f5e79ebc499c974098f3 | | LLaVA-1.5 | 32.3 | Not recorded | 6993f5e79ebc499c974098f3 | | GPT-4o | 32.3 | Not recorded | 6993f5e79ebc499c974098f3 | | LLaVA-N-Video | 31.2 | Not recorded | 6993f5e79ebc499c974098f3 | | mPLUG-Owl3 | 30.6 | Not recorded | 6993f5e79ebc499c974098f3 | | Qwen2-VL | 27.4 | Not recorded | 6993f5e79ebc499c974098f3 | | LLaVA-N-Inter | 24.2 | Not recorded | 6993f5e79ebc499c974098f3 | ### GPT-4o + ReViSe Narrative Accuracy: 51.9. Result ID: 699a7e315698665ec407b324. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"#Frames":"7.99","Param":"-"}. This record does not assert independent reproduction. ### VideoEspresso Narrative Accuracy: 45.2. Result ID: 699a7e315698665ec407b31d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"#Frames":"2.36","Param":"8.5B"}. This record does not assert independent reproduction. ### Qwen2-VL + ReViSe Narrative Accuracy: 39.2. Result ID: 699a7e315698665ec407b321. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"#Frames":"6.25","Param":"7B"}. This record does not assert independent reproduction. ### LongVA-DPO Narrative Accuracy: 35.5. Result ID: 699a7e315698665ec407b31a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"#Frames":"128","Param":"7B"}. This record does not assert independent reproduction. ### InternVL2 + ReViSe Narrative Accuracy: 35.5. Result ID: 699a7e315698665ec407b31f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"#Frames":"2.87","Param":"8B"}. This record does not assert independent reproduction. ### InternVL2 Narrative Accuracy: 33.9. Result ID: 699a7e315698665ec407b31e. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"#Frames":"FPS=1","Param":"8B"}. This record does not assert independent reproduction. ### Qwen-VL-Max Narrative Accuracy: 33.9. Result ID: 699a7e315698665ec407b322. [Reporting paper]() Recorded evaluation configuration: {"#Frames":"FPS=3","Param":"-"}. This record does not assert independent reproduction. ### LLaVA-1.5 Narrative Accuracy: 32.3. Result ID: 699a7e315698665ec407b318. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"#Frames":"4","Param":"7B"}. This record does not assert independent reproduction. ### GPT-4o Narrative Accuracy: 32.3. Result ID: 699a7e315698665ec407b323. [Reporting paper]() Recorded evaluation configuration: {"#Frames":"FPS=3","Param":"-"}. This record does not assert independent reproduction. ### LLaVA-N-Video Narrative Accuracy: 31.2. Result ID: 699a7e315698665ec407b31c. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"#Frames":"FPS=1","Param":"7B"}. This record does not assert independent reproduction. ### mPLUG-Owl3 Narrative Accuracy: 30.6. Result ID: 699a7e315698665ec407b31b. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"#Frames":"FPS=1","Param":"7B"}. This record does not assert independent reproduction. ### Qwen2-VL Narrative Accuracy: 27.4. Result ID: 699a7e315698665ec407b320. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"#Frames":"FPS=1","Param":"7B"}. This record does not assert independent reproduction. ### LLaVA-N-Inter Narrative Accuracy: 24.2. Result ID: 699a7e315698665ec407b319. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"#Frames":"FPS=1","Param":"7B"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.