| 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 | 31 |
Tags
- pedestrian detection
- ilvr
- fourmer
- daod
- neighboring pixel relationships
- ddrm
- facenet
- ssda-yolo
- unconditional ddpm
- ccdf
- conditional diffusion
- diffpir
- ddim
- ML
- unconditional generative models
- f2dnet
- ddpm
- linear inverse problem
- object detection
- semi-supervied learning
- diffusion
- deepfake detection
- IR
- stochastic contraction theory
- focal detection network
- face forgery detection
- Ai
- hqs-algorithm
- image restoration
- Triplet
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