- Reproduce Deep3dPortrait as baseline
- Improve expression manipulation and speed with SOTA 3DMM model
- Funded by Samsung Research
- Reproduce Noiseprint for blind deepfake detection
- Try multiple classification models like ResNet+RNN, C3D, and I3D
- Funded by NIS
- 3D Point Cloud needs sampling of large datapoint
- Try Classification, Registration, Restoration efficiently
- Try unsupervised learning for importance score like pruning
- Failed due to there is no concrete criteria like weight norm
- Network Pruning for Super Resolution
- Finding Lottery Ticket in Decoder (My Korean Blog Article)
- Try with multiple SR models like VDSR, MSRResNet, CARN, SRDenseNet, RDN, RRDB, RCAN
- Try baselines like SNIP, GraSP, and SynFlow
- Image classification with landmark location
- Try multiple models like EfficientNet, ResNet, ResNext, MoCo v2
- Get event prize money for sharing code
- NAVER AI Rush 2020 1 Round
- Advance to the next round with Rank 23
- Try multiple classification models like ResNet, EfficientNet
- Try multiple ways to deal with data imbalance like Focal Loss, Label Smoothing, Temperature Scaling
- Try multiple optimizers and schedulers like RAdam, CosineLRScheduler
Django-based Yonsei University Class Schedule Management Web Service
- Crawl class schedule and mileage history from Yonsei University Database
- Schedule editor like everytime
Gloves Hand recognition with OpenCV
Jingle Bell rings when hands are on each bell on the screen
