Visually Grounded Reasoning

Results across this thread

2 reported results from the papers in this thread.

How this thread developed

1. Oct 2026 · National University of Singapore, A*STAR Institute of Advanced Intelligence and Computing, Singapore

   Probability rewards keep visual grounding reasoning aligned with predicted objects

   Applies reinforcement learning with probability-based rewards to keep visual intention grounding chains aligned with the predicted object, mitigating thinking drift.

2. Oct 2026 · NVIDIA Research, The Hebrew University of Jerusalem

   Self-grounded attention improves visual reasoning across diverse benchmarks

   grounds reasoning steps in image regions via self-saliency to improve visual reasoning across benchmarks

3. Oct 2026

   Parallel diffusion decoding rivals autoregressive visual grounding at 4B scale

   Shows diffusion-based decoding can rival autoregressive grounding, broadening grounding model architectures.

4. Oct 2026

   Grounding foundation model surpasses larger vision-language models on precise perception and robotics

   Introduces a compact grounding model that outputs quantized coordinates, advancing precise visual grounding for downstream reasoning.

5. 2 further papers

   Oct 2026 · Peking University, The University of Melbourne

   Training video models across counterfactual views improves evidence-grounded answers

   Uses counterfactual video clips to improve evidence-grounded QA reasoning in video models.

6. 1 further paper

   Oct 2026 · University of Southern California, University of California San Diego

   Multimodal models can learn when impossible image edits require refusal

   Tests whether multimodal models can recognize impossible editing requests by grounding refusal in visual constraints, extending visually grounded reasoning to a refusal decision task.

7. 1 further paper

   Oct 2026 · University of California, Berkeley, Quotr AI

   Theory and experiments show when splitting large images into views helps VLMs find small objects

   provides theory for when multi-view splitting improves small object detection in VLMs.

   released code

8. 3 further papers

   Oct 2026 · Northwestern University, Carnegie Mellon University

   Shared visual transition training improves multimodal LLMs across diverse tasks

   Proposes a visual transition reasoning training source and benchmark that improves multimodal LLM performance on diverse tasks.

9. 2 further papers

   Oct 2026 · Ant Group, Tsinghua University

   Benchmark and agent for verifying images against real-world facts

   Introduces a benchmark and agent for verifying images against factual real-world knowledge, grounding reasoning in image regions.

10. 5 further papers

    Oct 2026 · York University, Macquarie University

    Learned attention adaptation improves language-guided video object segmentation

    Introduces an adapter for language-guided video object segmentation, grounding language in video regions without full model fine-tuning.

10 of 24 papers shown