cs.RONov 14, 2026

Semantics First, Pixels Later: Mission-Driven Event Streams for Drone Video over Bandwidth-Starved Links

Authors: Cong Xu

Organizations: Independent Researcher

Abstract

A surveillance drone on a bandwidth-starved link cannot send the video its operator would need, yet the operator's questions are often not about pixels: did three people converge, when, and where? We study what should cross a link of 1 Mbps down to 1 kbps instead, and in what order. BPS, a behavior-primitive stream, sends semantics first and pixels later: it runs a detector and a tracker onboard and sends short records (25 bytes on average) of multi-frame events selected by a mission card of 16-125 bytes, ahead of positions and pictures. We measure it by what a ground-side language model answers from it, under a preregistered protocol with a held-out set, against H.265 read by the vision fine-tune of the same model, H.265 on which the ground runs our own perception, a stream of quantized tracks, and onboard captions. On the development set (84% of questions about convergence, mostly one location) the event stream answers 74.3% of questions at 0.13 kbps. A reader that receives nothing answers 57.4%, and the video reader 59.6% at 100 kbps and 64.7% on the source frames. When the ground instead runs our perception on the video, video matches the stream at 100 kbps, about 800 times the stream's rate, and falls behind at 64 kbps and below. A track stream matches it at 96 times the bytes and loses 0.03 at 17 times. On held-out data the result is weaker: with events taken from annotated tracks, an oracle, the stream beats H.265 at 100 kbps (+0.13), but with onboard perception it does not (+0.01 [-0.04, +0.06]) and is no better than receiving nothing. Sent alone through emulated links the event records answer 0.739 at 1 Mbps and still 0.724 at 0.3 kbps; an every-frame track stream holds down to 1 kbps and video with ground-side perception down to about 100 kbps. On a Jetson Orin Nano the unchanged pipeline runs at 4.5 fps and misses most events.

Figures & tables

Appendix figures & tables11 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Apr 27, 2026

Don't Pause! Every prediction matters in a streaming video

Streaming video models should respond at the moment an event unfolds, not after the moment has passed. Yet existing online VideoQA benchmarks remain largely retrospective. They pause the video at fixed timestamps, pose questions about current or past events, and score models only at those moments. This protocol leaves streaming outputs untested. To close this gap, we introduce SPOT-Bench, featuring multi-turn proactive queries that evaluate general streaming perception required by an always-on, real-time assistant. SPOT-Bench comes with Timeliness-F1, a consolidated metric that measures streaming outputs by their temporal precision and balanced coverage across the entire video. Our benchmark reveals that during streaming inference: (i) MLLMs detect events reliably in a stream but spam predictions unprompted; (ii) post-training MLLMs for silence reduces spamming but induces unresponsiveness; (iii) half of the streaming video expects no response, which we term dead-time - compute spent here does not affect response latency. These findings motivate Ctrl-SPOT, a training-free streaming controller for MLLMs, that retains their event perception while controlling their streaming outputs for balanced coverage. This establishes a strong baseline on SPOT-Bench.
Oct 20, 2026

PACE: Plan, Attend, Check, and Emit for Training-Free Response Timing in Streaming Video Understanding

Streaming video understanding requires a model to decide not only what to answer but also when to answer. Evidence arrives progressively, and recent benchmarks identify premature responses, issued before the queried event has concluded, as a recurring failure mode. Existing approaches often learn response timing through additional training, while training-free alternatives typically rely on scalar gating signals that do not explicitly represent when an answer becomes final. We organize chunk-wise streaming errors into three patterns: sampling variation can make an answer unstable, a stable answer may lack support in the observed video, or an answer may be supported by the current evidence while the event remains unfinished. We propose PACE (Plan, Attend, Check, and Emit), a training-free framework designed around these three failure modes. Given a question, PACE drafts a plan specifying what to observe and what will signal completion. As the stream unfolds, it invokes an off-the-shelf video LLM at informative moments, evaluates each draft for stability, grounding in its cited frames, and completion, and records verified observations in an event ledger. Experiments on ProactiveVideoQA and OVO-Bench Forward show that PACE improves over the official protocols by up to 5.9 average PAUC points and 6.2 average accuracy points across proprietary and open-source backbones, yielding state-of-the-art performance.
Apr 2, 2026

A Simple Baseline for Streaming Video Understanding

Recent streaming video understanding methods increasingly rely on complex memory mechanisms to handle long video streams. We challenge this trend with a simple finding: a sliding-window baseline that feeds only the most recent N frames to an off-the-shelf VLM already matches or surpasses published streaming models. We formalize this baseline as SimpleStream and evaluate it against 13 major offline and online video LLM baselines on OVO-Bench and StreamingBench. Despite its simplicity, SimpleStream delivers consistently strong performance. With only 4 recent frames, it reaches 67.7% average accuracy on OVO-Bench and 80.59% on StreamingBench. Controlled ablations further show that the value of longer context is backbone-dependent rather than uniformly increasing with model scale, and reveal a consistent perception-memory trade-off: adding more historical context can improve recall, but often weakens real-time perception. This suggests that stronger memory, retrieval, or compression modules should not be taken as evidence of progress unless they clearly outperform SimpleStream under the same protocol. We therefore argue that future streaming benchmarks should separate recent-scene perception from long-range memory, so that performance improvements from added complexity can be evaluated more clearly.