Unsupervised video object-centric learning aims to decompose dynamic scenes into entity representations that persist over time. A new arXiv paper, "TSA: Temporal Slot Activation for Persistent Object-Centric Video Representation," proposes a method called TSA to address this task.

According to the abstract, existing recurrent video slot-attention methods propagate a fixed set of slots across frames. The abstract then cuts off mid-sentence at "but…", so the specific failure mode of those methods and the details of the TSA mechanism are not available in the source text.

Because the source is truncated, this digest can only report the problem framing and the proposed method's name. No experimental results or further technical claims are stated in the available text.