Research map / Vision-language and multimodal / Multimodal LLMs
SpatialSense: An Adversarially Crowdsourced Benchmark for Spatial Relation Recognition
Kaiyu Yang, Olga Russakovsky, Jia Deng
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| collect data with adversarial crowdsourcing instead of free annotation (ablation) | language-only model · SpatialNaive vs SpatialSense | accuracy: 69.2 -> 56.4 | Hurt |
| collect data with adversarial crowdsourcing instead of free annotation (ablation) | 2D-only model · SpatialNaive vs SpatialSense | accuracy: 71.3 -> 65.2 | Hurt |
| add 2D cues to language-only baseline (ablation) | language-only model · SpatialSense test | accuracy: 60.1 -> 71.1 | Improved |
| add language cues to 2D-only baseline (ablation) | 2D-only model · SpatialSense test | accuracy: 68.8 -> 71.1 | Improved |
| use state-of-the-art DRNet instead of 2D-only baseline | 2D-only model · SpatialSense test | accuracy: 68.8 -> 71.3 | Improved |
| train on SpatialSense instead of VRD/VG, evaluated cross-dataset | language-only model · cross-dataset predicate classification | accuracy drop: -1.8 | Improved |
| train on SpatialSense instead of VRD/VG, evaluated cross-dataset | 2D-only model · cross-dataset predicate classification | accuracy drop: -5.2 | Improved |
| test on SpatialSense-Positive instead of VG-Spatial | language-only model · predicate prediction from object pair | accuracy: 76.0 -> 39.8 | Hurt |
| test on SpatialSense-Positive instead of VG-Spatial | 2D-only model · predicate prediction from object pair | accuracy: 65.3 -> 43.4 | Hurt |
Closest papers by meaning
- Can Transformers Capture Spatial Relations between Objects? (ICLR 2024 · similarity 0.89)
- Spatial-DISE: A Unified Benchmark for Evaluating Spatial Reasoning in Vision-Language Models (ICLR 2026 · similarity 0.86)
- SpatialRGPT: Grounded Spatial Reasoning in Vision-Language Models (NeurIPS 2024 · similarity 0.86)
- SpatiaLab: Can Vision–Language Models Perform Spatial Reasoning in the Wild? (ICLR 2026 · similarity 0.85)
- Is A Picture Worth A Thousand Words? Delving Into Spatial Reasoning for Vision Language Models (NeurIPS 2024 · similarity 0.85)
- Stitch and Tell: A Structured Data Augmentation Method for Spatial Understanding (NeurIPS 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
