Find the hidden numbers behind characters
Published:
- Who: the solo- & side-project
- When: July 2020 - Aug. 2020
- Where: DACON Computer Vision Learning open competition for AI/ML, section in “Find the hidden numbers behind characters in EMNIST”.
- Dataset: Dacon dataset EMNIST
- What: Prediction the hidden numbers behind characters images
- How: TensorFlow 2.x. Implementation of ResNet121, DenseNet121, MobileNetV2, VGG19, and InceptionV3 from the scratch without pre-trained weights.
- Learn: SOTA models’ structures. Ensemble/Stacking method & Customized loss from winner codes. EDA and preliminary ML apporaches from the scratch.
- Through SOTA model evaluations without using pre-trained weights, I ranked the 21st-rank before 14 days to the deadline of submission, and I ranked at 49th of Private LB among 1185 participants. This is because I didn’t use the Ensemble/stacking models, and stopped submission by updating codes (due to family issues).
- The winner codes used the ensemble/stacking methods.
- The repository is opened to the public. My own codes were written in UCSD-Datahub server, but forgot to download it from there, so it was deleted.