| 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 |
Tags
- pedestrian detection
- face forgery detection
- stochastic contraction theory
- ddrm
- ddim
- daod
- Ai
- unconditional ddpm
- unconditional generative models
- ccdf
- semi-supervied learning
- neighboring pixel relationships
- diffusion
- facenet
- deepfake detection
- IR
- linear inverse problem
- f2dnet
- conditional diffusion
- ddpm
- ilvr
- ML
- focal detection network
- ssda-yolo
- Triplet
- image restoration
- fourmer
- diffpir
- object detection
- hqs-algorithm
Archives
- Today
- Total
목록DeepLearning (1)
Stand on the shoulders of giants
[paper review] Fast R-CNN (ICCV 2015)
Main IdeasFast R-CNN (Fast Region-based Convolutional Network method for object detection)mAP SOTA 달성 (R-CNN, SPPnet보다 우수한 성능)truncated svd를 사용하여 학습 시간은 줄이면서(Speed up 하면서) mAP를 거의 유지했다.RoI (max) pooling layerhxw ROI window를H x W grid가 되도록 h/H x w/W 크기의 sub windows로 나눈다.pooling을 각 feature map별로 진행spatial pytamid pooling layer의 원리(SPPnet)를 따르고 있다. Initializing from pre-trained networksFast R-CNN을 ..
Paper reviews
2024. 5. 6. 09:28