AI

KeyRec: Bounded Visual Memory for Streaming and Long-Video Understanding

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 05/10/2026, 18:18:24 UTC.

Briefings

KeyRec is a training-free framework for streaming and long-video understanding that constructs bounded visual memory by preserving recent observations in a cache and organizing historical evidence into an event bank. It uses a text-only router to allocate a fixed readout budget between recent and event memory without reprocessing historical frames. The method supports both modular encoder-projector VLMs and the encoder-projector-free NEO-ov architecture, achieving the best compressed performance in 13 of 15 settings using only 10% of the dense decoder-facing visual-token budget.

Key points

• KeyRec uses a text-only router to allocate a fixed readout budget between recent and event memory without reprocessing historical frames.
• KeyRec achieves the best compressed performance in 13 of 15 settings using only 10% of the dense decoder-facing visual-token budget.

Context and limitations

Results are limited to four streaming and long-video benchmarks across three VLM backbones, with specific performance metrics reported for NEO-ov 2B settings only.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 10/10/2026, 01:17:03 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeib56gejqqukonqxyv4ozhztxwfdltri2ngk2ghj2rec4ivnqxez5i