Topic provider

In the past 40 years, the Electronic and Optoelectronic System Research Laboratories undertakes the missions for the researches and developments of Semiconductor, Packaging, LED/OLED/Micro LED, Telecommunications, Flexible Electronics, 3D Imaging, Flexible Display and Transparent Display Application System technologies.

ITRI provides a welcoming work environment that stimulates the full potential of ITRI’s members, allowing them to excel and to perform their best in application-oriented researches. We fully cultivate the capability of independent innovation through participating in international cooperation and academic programs. These collaborative activities are pursued so as to aid in the success of technological incubation and entrepreneur start-ups. We strive to strengthen the technological innovation and value creation ability for Taiwan’s industry in order to remain the competitiveness globally.

Introduction

Automated optical inspection (AOI) [1] is an automated visual inspection of printed circuit board (PCB) (or LCD,transistor) manufacture where a camera autonomously scans the device under test for both catastrophic failure (e.g. missing component) and quality defects (e.g. fillet size or shape or component skew). It is commonly used in the manufacturing process because it is a non-contact test method. It is implemented at many stages through the manufacturing process including bare board inspection, solder paste inspection (SPI), pre-reflow and post-reflow as well as other stages. The Institute of Electronics and Optoelectronics in Industrial Technology Research Institute(ITRI) has spent years on developing flexible displays, hoping to elevate the production quality with AOI technology during the pilot run. This time we have invited experts from different fields to join us, and focus on identifying defect classifications of AOI image data that are offered so as to elevate the identification efficiency of AOI through statistical science.

This is an open-ended problem that can be solved consistently over a long period of time, and has offered milestone rewards for participants. Baseline is the best score of the previous milestone, and Private Leaderboard will be updated occasionally.

Reference
[1] https://en.wikipedia.org/wiki/Automated_optical_inspection

Activity time

The milestone is set at 23:59:59 on 11/04/2020, with its Private Leaderboard being announced at 00:00:00 on 11/05/2020. The grade would be viewed as a basis for awards. Private Leaderboard will be updated occasionally before the milestone.

During the period from 04/30 to 11/04, Private Leaderboard will be updated once two weeks.

Prize Information

The final score should be uploaded before 23:59:59 on the day of milestone. After model verification, the top three exceeding Baseline score (0.998521) on Private Leaderboard would be awarded as follows:

The first place: 100,000 discount points of hicloud

The second place: 100,000 discount points of hicloud

The third place: 100,000 discount points of hicloud

You will be awarded a certificate if your score outperforms Baseline on the Private Leaderboard and you pass model verification process

Rewards were provided by Chunghwa Telecom

Note:

1. Hicloud points can redeem service charge. After the discount, the price will be charged at 30% off based on the list price automatically.

2. Chunghwa Telecom deserves the right to make changes to the terms and conditions herein.

3. A model verification process will be performed to verify eligibility of awardees. Only those who pass model verification process will be awarded a certificate and/or prize.

Evaluation Criteria

After classification models of defect prediction are offered by researchers participating in the issue, the back-end of the system would process them in bathces regularly to calculate the score. Evaluations are conducted by calculating the corresponding accuracy rate of the actual value.

The following is the formula:

$$Accuracy = {\text{Number of correct predictions} \over \text{Number of total predictions}}$$

Data description

There are six categories included in image data that the issue offers (1 normal category + 5 defect categories)

• train_images.zip: image data for training (PNG format), 2,528 images in total.
• train.csv: includes 2 columns, ID and Label.
• ID: the image filename
• Label: defect classification category (0: normal, 1: void, 2: horizontal defect, 3: vertical defect, 4: edge defect, 5: particle)
• test_images.zip: image data for testing (PNG format), 10,142 images in total.
• test.csv: includes 2 columns, ID and Label.
• ID: the image filename
• Label: defect classification category (the value can be only one of the following numbers: 0, 1, 2, 3, 4, 5)

Please upload the file in CSV format, separate with commas, and save in one file. The content must correspond to number order of ID columns and include the following information.

• ID: the image filename
• Label: defect classification category (the value can be only one of the following numbers: 0, 1, 2, 3, 4, 5)

Rules

• Upload limit is 5 entries per day.
• The evaluation result would be based on the final version that you upload. If two teams (or more) got the same evaluation scores, the time they upload would determine the ranking.
• If there is cheating or fraud during the activity, the team that cheats will be disqualified from the activity and the vacancy would be filled up by other teams in the ranking order.
• This topic does not offer the team-up option by request of the topic provider. If there are answers identified the same during topic examination and audit, participants with identical answers would be forced to withdraw the activity, with all uploading data of the topic being eliminated.
• Keep the personal results in a safe place, lest the data be stolen or plagiarized, leading to the impairment of individual rights.
• Awarding: We consider an account to belong to an individual, thus the prize should be collected by only one person. Multiple winners of one award are no longer acceptable in the platform.
• After uploading, the answers of test data would be divided into two parts to calculate the score:
• Before the deadline of the activity: The system refers to partial Ground Truth of test data to examine and calculate the score. The result will be posted on Public Leaderboard as reference for the final score and ranking. This data accounts for 40% of the entire test data.
• After the deadline of the activity: The system refers to Ground Truth of the remaining test data to examine and calculate the score. The result will be posted on Public Leaderboard as reference for the final score and ranking.
• Any form of manual labeling is forbidden. A model verification process will be performed to verify eligibility of awardees. Only those who pass model verification process will be awarded a certificate and/or prize.