SimpleStream

Streaming video understanding methods keep adding memory and compression modules. SimpleStream shows a minimal sliding-window baseline that feeds only the most recent N frames to an off-the-shelf VLM matches or beats published methods.

It prompts a rethink of complex designs, with simple, reproducible code.

Features

Sliding window:Latest N frames.

Off-the-shelf:No retraining.

Strong results:Matches SOTA.