DetM-Tracker: Deterministic and Memory-Efficient VIS for Autonomous Industrial Inspection

Key points

• Autonomous industrial inspection requires strictly deterministic real-time inference under tight memory and computational constraints.1
• However, existing Video Instance Segmentation (VIS) models suffer from prohibitive scaling: dense temporal memory grows with spatial resolution and sequence length, while fixed-capacity decoders fail to exploit the inherent sparsity of industrial scenes, creating a practical memory wall on resource-constrained edge platforms.1
• DetM-Tracker eliminates architectural redundancy through three synergistic components: (1) Compact Scalar Memory, which condenses high-dimensional instance features into four-scalar descriptors, reducing temporal memory complexity from $O(T \cdot N \cdot R)$ to $O(T \cdot \hat{N})$.1
• Our approach effectively bridges the gap between modern VIS models and the practical demands of industrial edge deployment, reducing inference latency by approximately 20$\times$ and VRAM consumption by 6$\times$ compared to state-of-the-art dense-feature baselines.1

Sources (1)

• [1]DetM-Tracker: Deterministic and Memory-Efficient VIS for Autonomous Industrial Inspection
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 21, 05:09 AM

  We present DetM-Tracker, a streamlined VIS framework for autonomous industrial inspection, where target objects often exhibit highly similar appearance under constrained "micro" fields of view.

  Autonomous industrial inspection requires strictly deterministic real-time inference under tight memory and computational constraints.

Extractive summary: sentences quoted from the sources.