Genesis: Visual Augmentation Primitives for Multimodal Chain-of-Thought Reasoning
Authors: Fan Yang, Lifeng Chen, Yipu Wang, Han Wang, Feilong Chen, Yidong Ma, Zhiyang Chen, Chaoyang Zhao, Lingxiang Wu, Hanqi Jiang, Yibo Chen, Xu Tang, Yao Hu, Ming Tang, Xiaojie Jin, Xiaolong Zheng, Jinqiao Wang
Organizations: Foundation Model Research Center, Institute of Automation, Chinese Academy of Sciences · University of Chinese Academy of Sciences · Beijing Jiaotong University · State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences · Xiaohongshu Inc. · Independent Researcher · Wuhan AI Research
Recent multimodal large language models achieve promising results on visual reasoning tasks, approaching human-level performance on existing STEM benchmarks. However, these capabilities degrade under complex real-world scenes. We identify that chain-of-thought reasoning remains largely confined to the linguistic space, rarely anchoring each step in visual evidence. The main cause is the absence of such data during training. In this work, we introduce a training data augmentation pipeline, Visual Augmentation Primitives (VAP), which leverages state-of-the-art vision generative models to synthesize diverse photorealistic training samples and corresponding reasoning text. VAP introduces some basic operators and applies them to a seed image and gets a variant. It also constructs coherent multimodal reasoning chains throughout the generation process. Such training data force models to perceive visual content and perform logical reasoning according to them at every step. We further introduce multimodal chain-of-thought rubric-based reinforcement learning to enhance this. It rewards not only text reasoning but also visual reasoning by scoring intermediate reasoning steps for logical validity, completeness, and consistency. Moreover, we construct the UNREAL benchmark to systematically evaluate visual reasoning on hyperrealistic scenes across controllable difficulty levels. Experiments show that a 4B model built on this framework surpasses 8B baselines, with performance improving consistently as the model scales to 32B, validating the scaling law for visual reasoning.
Figures & tables
Figure 2: Two-stage data construction pipeline. Stage 1 (Surgical Answer-Changing Editing) : the Analyst extracts the global reasoning focus and partitions visual attributes into frozen and free dimensions, the Director defines an edit primitive, the Generator executes it, and the Verifier validates the output. Stage 2 (Axis Diversification & CoT Annotation) : the Planner selects free dimensions to diversify, the Generator produces visual variants, a four-step CoT is constructed for each, and the final dataset Dsyn is assembled.
Figure 3: Overview of multimodal chain-of-thought rubric-based reinforcement learning. (1) Rubric construction scores each reasoning step along three dimensions (logical validity, completeness, process consistency). (2) Reward aggregation combines format, rubric-based reasoning, and final-answer rewards, with final-answer correctness as the top priority.
Figure 4: UNREAL benchmark composition. The inner ring shows the five STEM families with their shares of the 600 samples. The outer ring shows the sixteen sub-domains with sample counts.
Model
ZeroBench
ZB-Sub
MathVista
MMMU
MMMU-Pro
EMMA
LogicVista
BabyVision
Closed-source models
GPT-5
2
-
81.3
84.2
78.4
-
71.8
-
Gemini-2.5-Pro
3
-
82.7
81.7
68.8
-
72.0
-
Claude-Opus-4.1
3
-
75.5
78.4
64.8
-
67.3
-
Open-source models
InternVL3-8B
-
-
71.6
62.7
-
-
44.1
-
Table 1: Main results on STEM visual reasoning benchmarks. Bold and underlined denote the best and second-best results among open-source models, respectively. Closed-source model results are shown for reference only and are excluded from ranking. Subscripts show the improvement over the corresponding Qwen3-VL baseline. “–” indicates a result that is still being verified.
Model
HallusionBench
MMStar
InfoVQA
Closed-source models
GPT-5
65.7
76.4
79.0
Gemini-2.5-Pro
63.7
77.5
84.2
Claude-Opus-4.1
60.4
72.1
69.4
Open-source models
InternVL3-8B
49.9
68.2
-
Table 2: Main results on VQA benchmarks. Bold and underlined denote the best and second-best results among open-source models, respectively. Closed-source model results are shown for reference only and excluded from ranking. Subscripts show the improvement over the corresponding Qwen3-VL baseline.
Model
Overall
Math.
Nat. Sci.
Percep.
Phys. & Eng.
Reas. & Log.
Closed-source models
Claude-Opus-4.7
32.9
28.6
46.8
28.3
32.4
41.2
Claude-Sonnet-4.6
30.7
23.4
45.2
32.6
35.3
36.1
Doubao-Seed-1.8
35.9
27.5
45.2
39.1
44.1
42.3
Gemini-3.1-Pro
37.1
23.2
51.6
43.5
53.9
44.3
GPT-5.5
59.6
62.9
51.6
54.3
56.9
62.9
Table 3: Results on the UNREAL benchmark, stratified by domain. Bold denotes the best result per column. Scores are accuracy (%).
Reward (32B)
ZeroBench
ZB-Sub
MathVista
MMMU
MMMU-Pro
EMMA
LogicVista
RLVR
3
28.44
82.20
75.22
66.42
45.23
66.22
Rubric
3
29.94
82.40
75.44
66.47
46.48
68.46
Table 4: Rubric-based RL vs. RLVR at 32B, holding data and optimizer constant.
RL Algorithm (4B)
ZeroBench
ZB-Sub
MathVista
MMMU
MMMU-Pro
EMMA
LogicVista
SAPO
0
20.96
77.2
63.39
52.20
34.61
59.06
GRPO
0
17.07
77.6
63.44
53.87
32.78
59.96
Table 5: Comparison of policy-optimization algorithms under rubric-based reward (4B).
Algebra, Geometry, Calculus, Statistics, Probability, Number Theory, Trigonometry, Linear Algebra, Combinatorics, Graph Theory, Topology, Differential Equations, Set Theory, Logic
Appendix
Table 6: STEM families and their sub-domains covered in our dataset.
Figure 5: Word cloud of question keywords in the training corpus. Word size is proportional to frequency. The distribution highlights the emphasis on quantitative reasoning (“number”, “value”, “total”, “calculate”), spatial understanding (“left”, “right”, “top”, “bottom”), and visual attributes (“color”, “shape”, “pattern”).
Category
Dimensions
Illumination
Lighting direction, intensity, color temperature
Viewpoint
Camera angle, distance, orientation
Object
Color, texture, shape, size, material
Background
Scene type, clutter level, color scheme
Layout
Object position, spatial arrangement, scale
Style
Rendering style, contrast, saturation
Appendix
Table 7: Visual taxonomy used for attribute decomposition. For each image, axes are partitioned into a frozen subset F (must be preserved pixel-wise to keep the test point intact) and a free subset V=A∖F that Stage 2 aggressively varies.
Figure 6: An example of a compositional visual reasoning dataset sample, where answering the question requires reasoning over object attributes such as color, shape, size, and material.
Figure 7: An example of a spatial visual reasoning dataset sample, where answering the question requires reasoning over relative object positions and counting objects satisfying spatial constraints.
Figure 8: An example of a visual counting dataset sample, where answering the question requires identifying and counting target objects in the scene.
Hyperparameter
4B
8B
32B
Learning rate
1×10−6
1×10−6
1×10−6
Train batch size
32
32
96
PPO mini-batch size
16
16
48
PPO micro-batch / GPU
4
4
1
Rollout group size n
8
8
8
Max prompt length
4096
8192
4096
Appendix
Table 8: Training hyperparameters at three model scales. The 4B and 8B variants share the same base configuration, while the 32B variant uses larger batch sizes, higher per-GPU token budgets, and multi-node training with rollout tensor parallelism.
Figure 9: Overview of the UNREAL benchmark construction pipeline. Top: the four sequential stages, namely (1) data collection and organisation across five STEM families and sixteen sub-domains, (2) LLM-driven question generation with MCQ → OE conversion, (3) structure-preserving image-to-image augmentation using the Axis Bank, and (4) human and automated quality control. Middle: the benchmark composition over the five STEM families together with format and scale statistics. Bottom: the key characteristics of the benchmark, including rule competition, structure preservation, high difficulty, mixed answer formats, and broad STEM coverage.
Figure 10: UNREAL benchmark illustration.
Family
Samples
Share
Mathematics
259
43.2%
Physics & Engineering
96
16.0%
Reasoning & Logic
92
15.3%
Perception
91
15.2%
Natural Science
62
10.3%
Total
600
100%
Appendix
Table 9: UNREAL benchmark composition by STEM family.
Family
Sub-domain
Samples
Mathematics
Combinatorial Geometry
63
Metric Geometry
40
Combinatorics & Counting
38
Descriptive Geometry
32
Solid Geometry
25
Transformation Geometry
23
Appendix
Table 10: UNREAL sub-domain distribution across the five STEM families.
Large Vision-Language Models (LVLMs) have shown strong promise for multimodal reasoning, yet often struggle with tasks requiring concepts beyond what is directly observable in the input image. Existing methods generate intermediate images or latent visual tokens to guide reasoning, but these representations can introduce errors and increasingly interfere with textual reasoning as reasoning progresses. We propose Visual-to-Text Chain-of-Thought Distillation (V2T), a framework that enables LVLMs to internalize visual reasoning without generating intermediate visual representations at inference time. V2T first trains a teacher LVLM using interleaved visual and textual chains of thought, and then uses knowledge distillation to train a student LVLM using the teacher's logits and cross-entropy supervision from ground-truth textual reasoning. When reasoning images can be mapped to the original image, V2T can additionally distill the teacher's attention to corresponding regions, while ground-truth bounding boxes can further guide a subsequent reinforcement learning stage. Experiments across multiple multimodal reasoning benchmarks show that V2T consistently outperforms the teacher and existing baselines, improving average accuracy by 14.3% on a held-out set and 2.7% on the broader visual evaluation suite. Moreover, lightweight SFT and substantially reduced RL make V2T up to 42x faster to train than state-of-the-art baselines.
Wenhan Yang, Nilay Naharas, Ali Payani +1
University of California, Los Angeles · Cisco Research
Chain-of-Thought (CoT) prompting has proven remarkably effective for eliciting complex reasoning in large language models (LLMs). Yet, its potential in multimodal large language models (MLLMs) remains largely untapped, hindered by the absence of large-scale datasets that capture the rich, spatially grounded reasoning intrinsic to visual understanding. Existing visual-CoT resources are typically small, domain-specific, or lack the human-like stepwise structure necessary for compositional visual reasoning. In this paper, we introduce VisReason, a large-scale dataset designed to advance visual Chain-of-Thought reasoning. VisReason comprises 489K annotated examples spanning four diverse domains, each featuring multi-round, human-like rationales that guide MLLMs through interpretable visual reasoning steps. Building upon this, we curate VisReason-Pro, a 165K subset produced with a stronger expert-level GPT annotator, enriched with detailed reasoning traces and 3D spatial grounding via depth-informed annotations. Fine-tuning the state-of-the-art Qwen2.5-VL model on VisReason and VisReason-Pro yields substantial improvements in step-by-step visual reasoning accuracy, interpretability, and cross-benchmark generalization. These results demonstrate that VisReason equips MLLMs with more systematic and generalizable reasoning capabilities. We envision VisReason as a cornerstone for cultivating human-like visual reasoning, paving the way toward the next generation of multimodal intelligence.
Lingxiao Li, Yifan Wang, Xinyan Gao +3
Boston University, Boston, MA, USA · Stony Brook University, Stony Brook, NY, USA · MMLab, The Chinese University of Hong Kong, Hong Kong SAR, China
Multimodal large language models (MLLMs) excel at visual reasoning but rely on text-based chain-of-thought (CoT), lacking interpretable visual intermediates. Existing methods use opaque tokens or external tools, missing key properties. We propose Gen-VCoT, a framework using expert vision models to generate RGB images as reasoning intermediates. It has three stages: visual grounding (SAM segmentation), geometric reasoning (Marigold depth maps), and semantic reasoning (Qwen2-VL integration). An adaptive router selects reasoning depth. Evaluations show Gen-VCoT improves spatial (25% better) and depth (50% better) questions, but may hurt simple factual queries. Text CoT outperforms visual intermediates on CLEVR (91.2% vs 62.5%), showing task-dependent optimal representations. Gen-VCoT establishes a new paradigm for interpretable multimodal reasoning.
Zhiqiang Zhou, Junliang Dai, Xu ling
Hunan Chemical Industry Vocational and Technical College, Zhuzhou, Hunan 412000, China