object recognition 2cvlab.khu.ac.kr/cvlecture20.pdf · 2018-11-27 · vgg net 4 vgg “very deep...
TRANSCRIPT
Object Recognition 2
Computer Vision
Alex Net
3
“ImageNet Classification with Deep Convolutional Neural Networks”, 2012
- ImageNet - ReLU, Dropout, Data augmentation
VGG Net
4
VGG “Very Deep Convolutional Networks for Large-Scale Image Recognition“. 2015
- ILSVRC 2014 Winner (localization and classification) - 3x3 conv, 2x2 pooling and deeper layers
GoogLeNet
5
GoogleNet “Going deeper with convolutions”, 2014
- ILSVRC 2014 - Inception modules, more than 100 layers
ResNet
6
MS ResNet “Deep Residual Learning for Image Recognition”, 2015
- ILSVRC 2015 Winner - Residual Block, 152 layers
R-CNN
7
R-CNN “Rich feature hierarchies for accurate object detection and semantic segmentation”
- Object detection including classification
Object Recognition in 3D
Computer Vision
https://www.youtube.com/watch?v=-l4ih6bpNTQ
https://www.youtube.com/watch?v=C4DO1nPJo1U
https://www.youtube.com/watch?v=lPQZIrdIo0g
https://www.youtube.com/watch?v=BjL-DW4jVEM
Introduction to the Literature
23
Top Conference in Computer Vision
1. CVPR
2. ICCV
3. ECCV
Top Journal in Computer Vision
1. IEEE TPAMI (IF: 9.455, JCR Ranking 2nd ~0.8%)
2. Springer IJCV (IF: 11.541, JCR Ranking 1st ~1.5%)
h-index: a scholar with an index of h has published h papers
each of which has been cited in other papers at least h times
https://scholar.google.com/citations?view_op=top_venues&hl=en&vq=eng_computervisionpatternrecognition
https://www.thecvf.com/ http://ieeexplore.ieee.org/
Oral presentation
Spotlight presentation
Poster presentation
CVPR 2018 conference
2 column < 9 pages
Peer Review by
AC & 3 Reviewers
~ 3300 submissions ~ 970 accepted papers
Final Project
Computer Vision
Topics 1. Stereo Matching Improvement
- 수업시간에 배운 방식으로 최적의 성능 내기 + Dynamic Programming 추가
- 기타 새로운 문제 해결 방식 추가: Occlusion, Hole filling, Boundary noise, thin object, non-textured region
Dataset: http://vision.middlebury.edu/stereo/data/scenes2014/
2. Panorama using Homography
- 신뢰도 높고 더 많은 Matching point pair 찾기: ex) SIFT, etc
- Homography 계산 정확도 향상, tone mapping
Dataset: use your own pictures (more than 10 images)
3. Change Detection
- 개선된 Detection Result
- 노이즈 제거, Clutter 제거, Morphological post procession
Dataset: http://jacarini.dinf.usherbrooke.ca/static/dataset/badWeather/skating.zip
4. CNN based Classification using Tensorflow and MNIST
- 개선된 Classification Result on distorted (affine, occlusion, noisy) test images
제출 내용: Commented Source Code, Result Images, Report Document
Pick one!!!!