Authors: Jia Li, Li Dai, Peng Jia, Zhenzhen Hu, Chee Seng Chan, Bingkun Bao, Richang Hong
Organizations: Hefei University of Technology Hefei, China · Hefei University of Technology Intelligent Interconnected Systems Laboratory of Anhui Province Hefei, China · Universiti Malaya Kuala Lumpur, Malaysia
In automated Printed Circuit Board Assembly (PCBA) inspection, standards-guided decisions require systems to jointly reason over fine-grained visual cues, component semantics, and manufacturing knowledge. Although large vision-language models (VLMs) provide a promising foundation, their deployment is hindered by the domain shift between standards-derived samples and real-world production-line imagery, together with heterogeneous output spaces spanning choice-based and numerical counting tasks. To address these challenges, we propose a multimodal reasoning framework for cross-domain PCBA visual question answering. The framework converts standards-derived, real-world, and auxiliary PCB-domain data into a unified instruction format and constructs verified reasoning traces aligned with visual evidence, question semantics, candidate options, and ground-truth answers. We further introduce Task-Aware Group Relative Policy Optimization (GRPO), which moves beyond exact-match supervision by integrating multi-component semantic rewards for choice-based questions, distance-aware rewards for counting questions, and an auxiliary format reward for valid outputs. During inference, answer-option semantic consistency correction, self-consistency voting, and multi-model arbitration are combined to improve prediction robustness. The proposed system achieves an Overall Score of 83.24 on the official PCBA Standard-to-Real Grand Challenge leaderboard, demonstrating the effectiveness of task-aware reward design and robust inference for cross-domain PCBA visual question answering.
Figures & tables
Figure 1. Illustration of cross-domain PCBA visual question answering, which combines standards-derived and real-world factory data and evaluates multimodal reasoning across factuality, cause and handling, quantitative reasoning, and attribute reasoning tasks.
Figure 2. The overall framework consists of two main stages: task-structured data construction and Task-Aware GRPO. In the first stage, standards-derived data, real-world data, and PCB-Bench data are converted into a unified multimodal instruction format, followed by task-specific CoT construction and verification. In the second stage, the policy model is optimized with GRPO using task-aware rewards, including choice-based rewards, counting rewards, and format rewards.
Rank
Submission
Overall
Attribute
Comp. Type
Count Comp.
Count Pin
Defect Exist.
Defect Type
Mount Side
Standard Know.
1
syds3sxhdjs
83.82
100.00
95.08
98.96
99.55
81.25
85.25
87.50
66.00
2
Ours
83.24
100.00
90.16
98.87
98.11
82.47
84.43
87.50
66.00
3
travis111
81.33
100.00
95.08
98.99
99.52
80.85
79.51
91.67
60.00
4
gmehong
80.11
100.00
96.72
98.96
98.13
76.92
77.05
91.67
61.00
5
dkhonker
77.52
100.00
93.44
98.72
97.58
77.08
73.77
75.00
58.00
Table 1. Comparison with the top five teams on the final PCBA Standard-to-Real Grand Challenge leaderboard. Our method is highlighted in gray and boldface. All scores are reported as percentages.
Setting
Overall
Attribute
Comp. Type
Count Comp.
Count Pin
Defect Exist.
Defect Type
Mount Side
Standard Know.
Backbone Scale and Auxiliary Data
Qwen3.5-4B
82.66
80.25
73.80
97.81
98.45
94.82
81.15
87.25
66.63
Qwen3.5-9B
83.99
98.75
83.90
97.84
97.66
93.91
77.75
84.50
68.75
Qwen3.5-27B
86.84
100.00
89.00
98.35
98.69
94.78
82.50
84.75
72.38
Qwen3.5-27B + PCB-Bench
86.93
99.50
90.10
97.55
98.89
95.09
82.05
84.50
72.69
GRPO Reward Design
Table 2. Ablation studies on the public evaluation set, evaluating backbone scale, auxiliary data, and GRPO reward design. All scores are reported as percentages, and the best result within each ablation group is shown in boldface.
Fine-grained visual reasoning remains challenging for vision-language models, especially when small but critical visual cues are buried in high-resolution images. Existing approaches rely on repeated cropping or test-time visual search to introduce local evidence, but they typically do not explicitly distinguish perception from reasoning. In this paper, we propose Perceive-to-Reason (P2R), a unified framework that formulates fine-grained visual reasoning as a two-stage process: the model first localizes question-relevant evidence as a Perceiver, and then answers the question as a Reasoner based on the annotated image and cropped regions. To better align training with this decoupled formulation, we further introduce Perception-Reasoning Alternating GRPO (PRA-GRPO), a role-aware reinforcement learning strategy that alternates between perception-focused and reasoning-focused updates using only final-answer supervision. Built on top of Qwen3-VL-Instruct-2B/4B/8B, P2R consistently improves performance across model scales. In particular, P2R-4B achieves 93.2% on V-Star, 81.9% on HR-Bench-4K, and 80.5% on HR-Bench-8K, substantially outperforming its corresponding backbone. Further experiments show that the benefits of P2R extend beyond high-resolution benchmarks to broader multimodal reasoning tasks. These results suggest that explicitly decoupling perception from reasoning provides an effective framework for fine-grained visual reasoning.
Multimodal large language models increasingly reason over screenshots and documents where the task itself may be written in pixels. Yet benchmarks usually place questions in text, leaving it unclear whether models use the same instruction equally well across channels. We introduce Visualized Task Semantics (VTS), a controlled intervention that moves the question into the image while keeping the source problem and answer fixed. Across six MLLMs and four benchmarks, accuracy drops in all 24 model-task pairs, by 17.8 points on average. Models often transcribe the visual question correctly yet fail to use it, exposing a semantic channel gap beyond OCR. To reduce this gap, we present prompt-region grounding, whose core design aligns the question region with typed semantics and recovers its clean representation from a masked view. At matched training cost, our method raises four-benchmark VTS accuracy from 58.0 to 66.3 while preserving accuracy on the original interface, and requires no OCR or region metadata at inference. Reading task-bearing text and grounding it as an instruction for reasoning are distinct capabilities.
Yongxin Wang, Ruizhe Zhou, Yueling Tang +4
Mohamed bin Zayed University of Artificial Intelligence · Sun Yat-sen University · Transsion +1
Establishing a clear link between model predictions and the visual evidence that supports them is critical for transparency and reliability in multimodal reasoning, yet current multimodal large language model (MLLM) evaluations do not explicitly enforce this alignment. Existing benchmarks assess either textual answer correctness or pixel-level localization in isolation, leaving the coupling of reasoning and grounding an open challenge. We introduce VISTAQA, a comprehensive benchmark for joint evaluation of free-form answer correctness and pixel-level evidence grounding in visual question answering. VISTAQA comprises 1,157 expert-curated samples spanning six task types and six visual domains, ranging from direct perception to compositional and relational reasoning. VISTAQA requires models to not only answer correctly, but to also provide precise segmentation masks that support their answers. It also includes hallucination-aware examples where no valid visual evidence exists. To support this enhanced evaluation, we introduce GROVE, a unified evaluation metric that enforces joint correctness by combining textual accuracy and grounding quality via a per-sample geometric mean, ensuring neither dimension can compensate for deficiencies in the other. Comprehensive experiments across grounding-aware models and hybrid pipelines with general-purpose MLLMs reveal that even the strongest systems achieve limited performance under GROVE, highlighting a substantial gap between answer accuracy and visual evidence alignment.
Mozhgan Nasr Azadani, Yimu Wang, Yongpeng Zhu +5
University of Waterloo · Stanford University · NVIDIA