FlowSG recasts scene graph generation as progressive flow matching on a hybrid discrete-continuous state using VQ-VAE tokens and a graph Transformer, delivering roughly 3-point gains over prior one-shot methods on VG and PSG.
Tical: Typicality- based consistency-aware learning for multimodal emotion recognition
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Can We Build Scene Graphs, Not Classify Them? FlowSG: Progressive Image-Conditioned Scene Graph Generation with Flow Matching
FlowSG recasts scene graph generation as progressive flow matching on a hybrid discrete-continuous state using VQ-VAE tokens and a graph Transformer, delivering roughly 3-point gains over prior one-shot methods on VG and PSG.