Sungmin Lee
Research scientist at the AIX Corporate R&D Center, SK Telecom, Korea. Focused on machine learning, computer vision, biomedical image processing and weakly supervised learning.
Latest Research Notes
Paper reviews, experiments, and technical insights.
Paper review
Instance segmentation
CenterMask : Real-Time Anchor-Free Instance Segmentation
[Summary]
Paper review
Instance segmentation
Swin Transformer: Hierarchical Vision Transformer using ShiftedWindows
[Summary]
Paper review
Instance segmentation
RetinaMask: Learning to predict masks improves state-of-the-art single-shot detection for free
[Summary]
Experiment
Quantization
Keras float32 model to tflite float16
Fig.1 float type에 따른 메모리 구조 비교 예시 (https://medium.com/@fanzongshaoxing/post-training-quantization-of-tensorflow-model-to-fp16-8d66b9dfa7...
Paper review
Classification
DeiT: Training data-efficient image transformers & distillation through attention
[Summary] 본 논문에서 제안하는 방법인 DeiT는 vision transformer (ViT)을 distillation으로 학습시킨 image classification 모델이다. DeiT의 main contribution은 ViT ...
Paper review
Machine Translation
Bridging the Gap Between Anchor-based and Anchor-free Detection via Adaptive Training Sample Selection
Introduction CVPR2020의 Best Paper Award List를 보던 중 제법 호기심을 자극하는 제목의 논문을 발견하여 읽게 되었다. 이 연구는 anchor-based 모델과 anchor-free model의 차이를 분석하고, ...