DeepIC Logo

DeepICLogo

Description The DeepIC Logo dataset detailed emerges a crucial response to the growing hardware challenges caused by undetected faults in printed circuit board (PCB) manufacturing. Its initial release contains 980 physical IC example images covering 119 unique IC manufacturer categories, for a total of 1010 logo instances. Of note, this version of the experiment uses only raw data derived from PCB optical images collected through the University of Florida SCAN laboratory facility, consistent with the FPIC data set, but specifically focused on annotating IC images with logos. Annotations include location details, logo orientation on the IC surface, and additional metadata such as IC type, manufacturer information (including country of origin), and operational status. An upcoming expanded version is designed to incorporate 3000 images containing approximately 1000 unique logo companies, including digital logos, and exclude variations present in IC package textures and physical IC samples. To address the challenge of class imbalance, the dataset admits that well-known companies exhibit more instances than less well-known manufacturers.
Downloads DeepICLogo
Citations
×

Citation Formats

Text Format BibTeX
S. Ghosh, P. Craig, J. Julia, N. Varshney, H. Dalir and N. Asadizanjani, DeepICLogo: A Novel Benchmark Dataset for Deep Learning-Based IC Logo Detection, 2023 IEEE Physical Assurance and Inspection of Electronics (PAINE), Huntsville, AL, USA, 2023, pp. 1-8. @INPROCEEDINGS{10318008, author={Ghosh, Shajib and Craig, Patrick and Julia, Jake and Varshney, Nitin and Dalir, Hamed and Asadizanjani, Navid}, booktitle={2023 IEEE Physical Assurance and Inspection of Electronics (PAINE)}, title={DeepICLogo: A Novel Benchmark Dataset for Deep Learning-Based IC Logo Detection}, year={2023}, volume={}, number={}, pages={1-8}, doi={10.1109/PAINE58317.2023.10318008}}