AI

LVMT: Video Mask Transformer for Long-term Video Segmentation

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 05/10/2026, 09:18:24 UTC.

Briefings

The Long-term Video Mask Transformer (LVMT) addresses long-term video segmentation by combining a lightweight GRU-based temporal propagation module with Truncated Query Propagation (TQP). This approach enables training on long videos by processing frames in chunks where object information propagates between chunks but backpropagation occurs only within individual chunks, solving memory and gradient issues. Extensive experiments on six benchmarks demonstrate LVMT as a new state-of-the-art, achieving 10x speed compared to prior methods while maintaining efficiency. The research was published in September 2026.

Key points

• The Long-term Video Mask Transformer (LVMT) introduces Truncated Query Propagation (TQP), a training strategy that processes video frames in chunks to enable long-term video segmentation without out-of-memory issues or vanishing gradients.
• The model incorporates a lightweight GRU-based temporal propagation module that learns to selectively select and propagate object information across time, addressing limitations in existing online video segmentation methods.

Context and limitations

The study covers six video segmentation benchmarks; specific results for each task are not detailed. The research is presented as a preprint with no formal peer review or publication in a journal.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 05/10/2026, 09:31:36 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeicuxbiekmnkfk2y7bus3x43sz3gnh45452il7blh65mti7bdhmqg4