# Frame Selection for long-form video QA on 10-minute video 600 frames at 1 FPS, K=16 Source: [Frame Selection for long-form video QA on 10-minute video 600 frames at 1 FPS, K=16]() Collection index: [llms.txt]() Data updated: 2026-03-20T09:19:25.035Z Evaluation of models designed for frame selection within long-form videos (specifically 10 minutes, 600 frames total at 1 FPS) to support Question Answering tasks. Metrics focus on end-to-end latency and the interpretability of the selected frames for compositional questions. 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 | Frame selection for long-form video QA | | Dataset (including any stated split) | 10-minute video 600 frames at 1 FPS, K=16 | | Metric | E2E Latency (s) | | Unit | TIME | | Direction | Lower 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 - [E2E Latency (s)](): Lower is better; unit: TIME. - [Amortized Latency (s)](): Lower is better; unit: TIME. - [Interpretability (Compositional Qs)](): Higher is better; unit: SCALAR. ## Recorded results | Method | E2E Latency (s) | Result date | Report paper | | --- | --- | --- | --- | | FFS | 0.1 | Not recorded | 69bcd4760636e5bf4afc296a | | VidF4 | 1.1 | Not recorded | 69bcd4760636e5bf4afc296a | | Frame-Voyager | 1.6 | Not recorded | 69bcd4760636e5bf4afc296a | | MDP3 | 1.8 | Not recorded | 69bcd4760636e5bf4afc296a | | AKS | 2.7 | Not recorded | 69bcd4760636e5bf4afc296a | | BOLT | 3 | Not recorded | 69bcd4760636e5bf4afc296a | | NeuS-QA | 6.7 | Not recorded | 69bcd4760636e5bf4afc296a | | T\* | 13 | Not recorded | 69bcd4760636e5bf4afc296a | | VSLS | 13.3 | Not recorded | 69bcd4760636e5bf4afc296a | | HiMu | 13.3 | Not recorded | 69bcd4760636e5bf4afc296a | | VideoZoomer | 16 | Not recorded | 69bcd4760636e5bf4afc296a | | VideoAgent | 44.7 | Not recorded | 69bcd4760636e5bf4afc296a | | SeViLA | 60 | Not recorded | 69bcd4760636e5bf4afc296a | ### FFS E2E Latency (s): 0.1. Result ID: 69bd110b47f34ce924a76026. [Reporting paper]() Recorded evaluation configuration: {"Train-Free":false,"Query Representation":"Learned policy","Selection Evidence":"Vis."}. This record does not assert independent reproduction. ### VidF4 E2E Latency (s): 1.1. Result ID: 69bd110b47f34ce924a76027. [Reporting paper]() Recorded evaluation configuration: {"Train-Free":false,"Query Representation":"Learned score fn.","Selection Evidence":"Vis."}. This record does not assert independent reproduction. ### Frame-Voyager E2E Latency (s): 1.6. Result ID: 69bd110b47f34ce924a76025. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Train-Free":false,"Query Representation":"Learned score fn.","Selection Evidence":"Vis."}. This record does not assert independent reproduction. ### MDP3 E2E Latency (s): 1.8. Result ID: 69bd110b47f34ce924a7601e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Train-Free":true,"Query Representation":"Global embedding","Selection Evidence":"Vis."}. This record does not assert independent reproduction. ### AKS E2E Latency (s): 2.7. Result ID: 69bd110b47f34ce924a7601d. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Train-Free":true,"Query Representation":"Global embedding","Selection Evidence":"Vis."}. This record does not assert independent reproduction. ### BOLT E2E Latency (s): 3. Result ID: 69bd110b47f34ce924a7601c. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Train-Free":true,"Query Representation":"Global embedding","Selection Evidence":"Vis."}. This record does not assert independent reproduction. ### NeuS-QA E2E Latency (s): 6.7. Result ID: 69bd110b47f34ce924a76021. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Train-Free":true,"Query Representation":"Temporal logic spec.","Selection Evidence":"Vis."}. This record does not assert independent reproduction. ### T\* E2E Latency (s): 13. Result ID: 69bd110b47f34ce924a7601f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Train-Free":true,"Query Representation":"Flat object queries","Selection Evidence":"Vis. (OVD)"}. This record does not assert independent reproduction. ### VSLS E2E Latency (s): 13.3. Result ID: 69bd110b47f34ce924a76020. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Train-Free":true,"Query Representation":"Fixed relation triplets","Selection Evidence":"Vis. (OVD)"}. This record does not assert independent reproduction. ### HiMu E2E Latency (s): 13.3. Result ID: 69bd110b47f34ce924a76028. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Train-Free":true,"Query Representation":"Hierarchical logic tree","Selection Evidence":"Vis.+Aud."}. This record does not assert independent reproduction. ### VideoZoomer E2E Latency (s): 16. Result ID: 69bd110b47f34ce924a76024. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Train-Free":false,"Query Representation":"Implicit (iterative VLM)","Selection Evidence":"Vis."}. This record does not assert independent reproduction. ### VideoAgent E2E Latency (s): 44.7. Result ID: 69bd110b47f34ce924a76022. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Train-Free":false,"Query Representation":"Implicit (iterative LLM)","Selection Evidence":"Vis."}. This record does not assert independent reproduction. ### SeViLA E2E Latency (s): 60. Result ID: 69bd110b47f34ce924a76023. [Reporting paper]() Recorded evaluation configuration: {"Train-Free":false,"Query Representation":"Implicit (per-frame VLM)","Selection Evidence":"Vis."}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.