Call for Papers
The CogVL workshop provides a forum for students and researchers across computer vision, natural language processing, and cognitive science to explore how cognitively inspired frameworks can enhance the robustness, generalization, and interpretability of VLMs. The workshop will feature invited talks by leading experts and a panel discussion examining how cognitive principles can reshape model architectures, learning paradigms, and evaluation methodologies.
Submission Tracks
We welcome submissions covering technical contributions, evaluations, and position papers. Submissions may address topics pertaining to cognitively inspired VLMs, including, but not limited to the topics listed below.
We will have two tracks for paper submissions:
1 Track 1: Papers with IEEE/CVF Workshop Proceedings
Page limit: 8 pages (excluding references)
Requirements:
- Original, previously unpublished papers
- Dual submissions are not allowed
- Accepted papers will be included in the IEEE/CVF workshop proceedings
2 Track 2: Papers without Workshop Proceedings
Page limit: ≤ 8 pages (excluding references)
Requirements:
- Papers will not be included in the proceedings
- Accepted papers will be publicly shared on the workshop website
Two subcategories:
- Novel/ongoing work: Limited to 4 pages (excluding references)
- Accepted/previously published papers: Limited to 8 pages (excluding references)
Submission Site
Submit your papers via OpenReview: https://openreview.net/group?id=thecvf.com/CVPR/2026/Workshop/CogVL
Topics of Interest
1 Robustness and Generalization
How can cognitive principles help models overcome spurious correlations and shortcut learning? What evaluation paradigms can reveal whether models have developed robust, generalizable understanding?
Example topics: Methods for reducing reliance on superficial features through structured priors, benchmarks probing out-of-distribution generalization, training paradigms inspired by developmental psychology or curiosity-driven learning, and interpretability analyses revealing cognitive-like reasoning patterns versus shortcut strategies.
2 Causal and Counterfactual Reasoning
How can VLMs move beyond surface correlations to understand causal relationships in visual scenes? What role does counterfactual thinking play in robust multimodal reasoning?
Example topics: Methods for learning causal world models from visual data, architectures supporting counterfactual inference and intervention, benchmarks evaluating causal understanding in image and video, and position papers on the role of causal reasoning.
3 Compositional and Structured Reasoning
How can models learn to compose visual concepts and reason over structured representations? What cognitive principles can guide the development of compositional generalization in VLMs?
Example topics: Architectures for compositional visual reasoning, methods for learning object-centric or scene graph representations, benchmarks testing systematic generalization and compositional understanding, and analyses of how compositional biases emerge during training.
4 Theory of Mind and Social Reasoning
How can VLMs develop an understanding of agents' mental states, goals, and intentions? What is required for genuine social reasoning in multimodal contexts?
Example topics: Methods for inferring goals and intentions from visual observations, benchmarks evaluating theory of mind in social scenarios, approaches to modeling perspective-taking and false belief understanding, and position papers on the role of social cognition in embodied AI.
5 Non-Monotonic Reasoning
How can models infer the most likely explanations for observed phenomena? How can they revise their conclusions when presented with new evidence?
Example topics: Methods for hypothesis generation and explanation, approaches to belief revision and non-monotonic reasoning, benchmarks for evaluating reasoning under uncertainty and surprise, and analyses of how models handle unexpected or anomalous observations.
6 Dual-process Reasoning and Meta-Cognition
How can we bridge fast, intuitive processing (System 1) with slow, deliberate reasoning (System 2) in VLMs? What role does meta-cognitive monitoring play in reliable reasoning?
Example topics: Architectures supporting both reflexive and reflective processing, methods for adaptive computation and reasoning, approaches to uncertainty estimation and confidence calibration, computational principles underlying dual-process theories.
7 Cognition for Embodied and Interactive Agents
How can cognitively inspired approaches improve situational awareness, planning, and collaboration in embodied AI systems? What cognitive capabilities are essential for human-robot interaction?
Example topics: Methods for goal inference and intention recognition in interactive settings, approaches to collaborative reasoning and shared mental models, and cognitive architectures for embodied agents.
Important Dates
- March 1, 2026 - Workshop Paper Submission Deadline
- March 25, 2026 - Notification to Authors
- April 5, 2026 - Camera Ready Deadline
Review Process
Poster Presentations
All accepted papers are invited to present at the poster session, fostering in-depth discussion and cross-disciplinary exchange with attendees.
Formatting Requirements
Papers should follow the CVPR 2026 formatting guidelines. Page limits vary by track (see Submission Tracks above). All submissions must be in PDF format.