티스토리 뷰
요새 이미지로부터 Dense Prediction을 해야하는 모델을 개발중에 있다.
그래서 이분야의 초석이라 할 수 있는 U-Net모델을 베이스라인으로 삼아서 테스트 해보고있다.
https://arxiv.org/abs/1505.04597
U-Net: Convolutional Networks for Biomedical Image Segmentation
There is large consent that successful training of deep networks requires many thousand annotated training samples. In this paper, we present a network and training strategy that relies on the strong use of data augmentation to use the available annotated
arxiv.org

무려 2015년에 나온 논문이였다.
무려 내가 학부 2학년때 나온 논문이다...
하도 유명한 네트워크고 Segmentation이라는 Classification 보다 고수준의 Task다보니 어렴풋이 생각했을때 어느정도 Classification Task 에서 많은 발전이 있고나서(내 생각엔 ResNet 나오고나서) 나온 줄 알았는데 그게 아니라 살짝 충격을 받았다.
우선 아래 프로젝트에서 모델 구조 그대로 갖다가 from scratch로 써보고 있다.
https://github.com/milesial/Pytorch-UNet
GitHub - milesial/Pytorch-UNet: PyTorch implementation of the U-Net for image semantic segmentation with high quality images
PyTorch implementation of the U-Net for image semantic segmentation with high quality images - milesial/Pytorch-UNet
github.com
'Deep Learning' 카테고리의 다른 글
2025년 2월 22일 기준 최신 vision encoder! (0) | 2025.02.22 |
---|---|
pytorch docker 이미지에서 onnxruntime-gpu cuda 인식 안되는 경우 (0) | 2025.02.21 |
다양한 최신 백본으로 Image Classification 성능 측정한 논문 - 다시 보기용 (0) | 2025.02.19 |
torch2onnx transformer encoder, decoder onnx변환시 aten::_native_multi_head_attention 에러 우회법 (0) | 2025.02.18 |
face recognition task와 face alignment (0) | 2025.01.31 |
- Total
- Today
- Yesterday
- ㅂ
- 단축키
- PyCharm
- 순열
- 인공지능을 위한 선형대수
- MOT
- LCA
- 가장 긴 증가하는 부분 수열
- 백트래킹
- 백준
- cosine
- 조합
- 이분탐색
- 백준 11437
- 자료구조
- 문제집
- 파이참
- 위상 정렬 알고리즘
- 백준 1766
- FairMOT
- Lowest Common Ancestor
- 백준 11053
- C++ Deploy
일 | 월 | 화 | 수 | 목 | 금 | 토 |
---|---|---|---|---|---|---|
1 | 2 | 3 | 4 | 5 | ||
6 | 7 | 8 | 9 | 10 | 11 | 12 |
13 | 14 | 15 | 16 | 17 | 18 | 19 |
20 | 21 | 22 | 23 | 24 | 25 | 26 |
27 | 28 | 29 | 30 |