Reproduce “Semantic-Aware Neural Video Codec for Error-Resilient Low-Latency Transmission” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Matin Mortaheb, Homa Esfahanizadeh, Jinfeng Du, Harish Viswanathan
A semantic-aware extension of the DCVC-RT neural video codec is designed for low-latency transmission over packet-erasure channels. By combining independent packet decoding, spatial–feature latent partitioning, and learned unequal error protection, the system avoids error propagation and preserves query-relevant regions more effectively than uniform-protection decoding under both uniform and heterogeneous erasures.
Key insights
- Hybrid spatial–feature packetization enables unequal protection of both important regions and informative latent channels, unlike feature-only packet grouping.
- Semantic weighting is incorporated into both reconstruction training and packet assignment, allowing the codec to trade quality in less relevant areas for improved preservation of task-critical content.
- Independent entropy modeling is a structural robustness mechanism: packet losses affect only the corresponding latent portions instead of propagating through the bitstream.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.