SPARROW: Learning Spatial Precision and Temporal Referential Consistency in Pixel-Grounded Video MLLMs Paper • 2603.12382 • Published Mar 12 • 2