Characterize practical edge-device deployment impact

Characterize the practical impact of device-side compute constraints, communication overhead, and network-latency variability for the Caption-once, Frames-on-Demand framework on representative edge hardware.

Background

The Caption-once, Frames-on-Demand framework is evaluated using a simulated separation between offline edge-side indexing and online cloud-side reasoning, rather than on an actual mobile, wearable, or other resource-constrained edge device. The paper notes that its default Qwen3-VL-8B captioner assumes a modest server-class device or powerful desktop, and that deployment on smartphones or smart glasses would require additional model compression. Consequently, the effects of realistic device computation, communication, and latency conditions remain unresolved and require systematic measurement on representative hardware.

References

Finally, our edge-cloud setup is validated under simulated separation rather than physical deployment, so the practical impact of device-side compute constraints, communication overhead, and network latency variability remains to be systematically characterized on representative edge hardware.

— Caption-once, Frames-on-Demand: Visual-Need Routing for Budget-Aware Agentic Long Video Understanding  (2609.11899 - Cai et al., 10 Sep 2026) in Limitations, subsection “Definition of ‘edge’”