MindView benchmark tests generative multi-view spatial intelligence
Original title: MindView: Benchmarking Generative Multi-View Spatial Intelligence
Fine-tuning BAGEL on MindView raises its overall score by 24.3 points and improves all seven tasks.
MindView contains 1,500 samples from 513 scenes across three indoor datasets and seven tasks in geometry-dominant and semantics-dominant families. Evaluation of eleven generators shows substantial room for improvement, and all eleven score lower under expanded context. Fine-tuning BAGEL on the MindView training set improves generation and also raises scores on three spatial-understanding benchmarks without their annotations. Training-free MindCue reverses the expanded-context score drop.
Unconfirmed: Generator names besides BAGEL are not listed; scores are the authors' evaluations.
- The benchmark has 1,500 samples from 513 scenes in three indoor datasets.
- All eleven generators obtain lower overall scores under expanded context.
- Fine-tuned BAGEL gains 24.3 overall points across all seven tasks.
- MindCue is a training-free selector guided by a target-view hypothesis.
- AI models
- BAGEL
From the original article
1 min read at arxiv.orgSpatial intelligence has recently been assessed through generative tasks such as spatially grounded image editing, yet language-guided generation of spatially consistent views from partial multi-view observations remains underexplored. We introduce MindView, a benchmark for generative multi-view spatial intelligence with 1,500 samples from 513 scenes across three indoor datasets.
By Yanbo Wang · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.