Presents SpatialScore benchmark for MLLM spatial reasoning, evaluates 49 models showing large human gap, and supplies SpatialCorpus plus SpatialAgent to improve performance.
Canonical reference
Sims-v: Simulated instruction-tuning for spatial video understanding
Canonical reference. 100% of citing Pith papers cite this work as background.
5
Pith papers citing it
Background
100% of classified citations
citation-role summary
background 5
citation-polarity summary
roles
background 5polarities
background 5representative citing papers
Adding per-frame camera-pose supervision to a video MLLM improves spatial and general video question answering by 2–6% and yields SOTA streaming pose estimates on ScanNet.
MolmoAct2 is an open VLA model that outperforms baselines like Pi-05 on 7 benchmarks and whose backbone surpasses GPT-5 on 13 embodied-reasoning tasks through new datasets, specialized training, and architecture changes for lower latency.
Unlabeled web videos processed by designed data engines generate effective training data that yields strong zero-shot and finetuned performance on 3D detection, segmentation, VQA, and navigation.
LychSim introduces a controllable simulation platform on Unreal Engine 5 with Python API, procedural generation, and LLM integration for vision research tasks.
citing papers explorer
-
SpatialScore: Towards Comprehensive Evaluation for Spatial Intelligence
Presents SpatialScore benchmark for MLLM spatial reasoning, evaluates 49 models showing large human gap, and supplies SpatialCorpus plus SpatialAgent to improve performance.
-
Cambrian-P: Pose-Grounded Video Understanding
Adding per-frame camera-pose supervision to a video MLLM improves spatial and general video question answering by 2–6% and yields SOTA streaming pose estimates on ScanNet.
-
MolmoAct2: Action Reasoning Models for Real-world Deployment
MolmoAct2 is an open VLA model that outperforms baselines like Pi-05 on 7 benchmarks and whose backbone surpasses GPT-5 on 13 embodied-reasoning tasks through new datasets, specialized training, and architecture changes for lower latency.
-
Lifting Unlabeled Internet-level Data for 3D Scene Understanding
Unlabeled web videos processed by designed data engines generate effective training data that yields strong zero-shot and finetuned performance on 3D detection, segmentation, VQA, and navigation.
-
LychSim: A Controllable and Interactive Simulation Framework for Vision Research
LychSim introduces a controllable simulation platform on Unreal Engine 5 with Python API, procedural generation, and LLM integration for vision research tasks.