TSG Bench, a new benchmark, reveals that LLMs handle scene graph understanding well but perform poorly at generating scene graphs from complex narratives, with action decomposition as the main bottleneck.
- Do not output any additional text or explanation
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
LLM Meets Scene Graph: Can Large Language Models Understand and Generate Scene Graphs? A Benchmark and Empirical Study
TSG Bench, a new benchmark, reveals that LLMs handle scene graph understanding well but perform poorly at generating scene graphs from complex narratives, with action decomposition as the main bottleneck.