PCBA_Standard-to-Real_Challenge Access Request
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
Please fill in the following information to request access to the dataset. If you are a student, it is recommended that you fill in the name of your supervising professor.
Log in or Sign Up to review the conditions and access this dataset content.
Here is a more concise, English version of the Hugging Face dataset card. You can copy and paste this directly into your README.md.
Dataset Card for PCBA Standard-to-Real Challenge
🏆 Challenge Overview
This is the official dataset for the PCBA Standard-to-Real Grand Challenge, held in conjunction with ACM Multimedia (MM) 2026.
This challenge focuses on Cross-domain Visual Question Answering (VQA) for real-world manufacturing inspection. The goal is to develop multimodal models capable of generalizing from theoretical manufacturing standards (normative illustrations) to dense, heterogeneous real-world Printed Circuit Board Assembly (PCBA) imagery to detect defects, reason about causes, and make actionable decisions.
🏆 Competition Platform
Participants can register for the challenge, access submission instructions, and submit prediction files through the competition platform. High-performing teams will be considered for organizer recommendation for the ACM Multimedia 2026 Grand Challenge paper submission, subject to the challenge rules and conference requirements.
- Competition Platform: PCBA Standard-to-Real Grand Challenge on Codabench
📊 Data Splits
| Standard | Factuality | Quantitative | Attribute | Total | |||||
|---|---|---|---|---|---|---|---|---|---|
| Component Type | Mount Side | Defect Existence | Defect Type | Component | Pin/Lead | ||||
| Train | 1600 | 1000 | 400 | 2000 | 2000 | 400 | 400 | 400 | 8200 |
| Test | 1600 | 1000 | 400 | 2000 | 2000 | 400 | 400 | 400 | 8200 |
| Total | 3200 | 2000 | 800 | 4000 | 4000 | 800 | 800 | 800 | 16400 |
🎯 VQA Taxonomy
Models are evaluated across four distinct VQA dimensions:
- Standard-based Knowledge QA: Interpreting standard documents, analyzing defect causes, and making handling decisions.
- Factuality: Component type recognition, mount-side identification, defect existence, and defect type detection.
- Quantitative Reasoning: Counting components and pins/leads.
- Attribute Reasoning: Recognizing the shape and visual attributes of main components.
🔗 Resources & Links
- Challenge Homepage: PCBA Standard-to-Real Grand Challenge
- Competition Platform: Codabench Competition
- Contact: aimmifm@gmail.com
- Downloads last month
- 899