ResearchBenchmarksScene & action recognition on SUN397 T=50Follow83.55Last-Task Accuracy (AT)VILA79.372581.4612583.5585.63875Feb 14, 2026Evaluation ResultsMethodMethodLinksLast-Task Accuracy (AT)Average Incremental Accuracy (A_bar)VILABackbone=ViT-B/16Backbone=ViT-B/162026.0283.5589.53