CV Chapter 6 Categorization 2
Questions about the lecture 'Computer Vision' of the RWTH Aachen Chapter 6 Categorization 2
Questions about the lecture 'Computer Vision' of the RWTH Aachen Chapter 6 Categorization 2
Set of flashcards Details
| Flashcards | 50 |
|---|---|
| Language | English |
| Category | Computer Science |
| Level | University |
| Created / Updated | 04.02.2017 / 21.02.2017 |
| Weblink |
https://card2brain.ch/box/20170204_cv_chapter_6_categorization_2
|
| Embed |
<iframe src="https://card2brain.ch/box/20170204_cv_chapter_6_categorization_2/embed" width="780" height="150" scrolling="no" frameborder="0"></iframe>
|
How are the weights of bbox reg of R-CNN computed? [5]
1. wTc,xxpool5 = x*-x / w
2. wTc,yxpool5 = y*-y / h
3. wTc,wxpool5 = ln(w*/w)
4. wTc,hxpool5 = ln(h*/h)
5. With xpool5 features from pool5 layer
What are the problems of R-CNN? [5]
1. Fine tune network with softmax classifier (log loss)
2. Train post-hoc linear SVMs (hinge loss)
3. Train post-hoc bounding box regressors (squared loss)
4. Long and slow training (3 days and 47s per image)
5. High memory consumption (Easily 200GB)
What are the techniques of fast R-CNN? [2]
1. Forward and 2. backward pass
What are the characteristics of region proposal networks (RPN) for CNN? [4]
1. Remove dependence on external region proposal algorithm
2. Get region proposal from same CNN
3. Use feature sharing
4. Single pass object detection becomes possible
What is the definition of faster R-CNN?
Fast R-CNN + RPN
What are the losses of proposals and RoI pooling for faster R-CNN? [3]
1. Total 4
2. Classification loss
3. Bounding box regression loss
What is the definition of fully convolutional networks (FCNs)? [4]
1. All operations formulated as convolutions
2. Can process arbitrarily sized images
3. Forward with inference
4. Backward with learning
How does semantic image segmentation works with FCNs? [3]
1. Perform pixel-wise prediction
2. Sliding-window classification producing heatmap of scores
3. Avoid low resolution with up-sampling and skip connections
How does human pose estimation works with FCNs? [2]
1. Choose key-points with target disk (r) for skeleton joints
2. Each disk has ground-truth label of 1
How does face verification with embedding space works with FCNs? [3]
1. Use triplet network with negatives, anchors and positives
2. Learn function grouping positives closer to anchors
3. Vector arithmetics possible due to linear regularities
