Object detection

Question 1

What is object detection?

Answer 1

Object detection detects where in the image we have objects using a bounding box and classifies them

Question 2

How can we do object detection using sliding windows=

Answer 2

Do a classification on every sliding window, resulting in classification as a object or background.

Question 3

Why don’t we usually use the sliding window approach for object detection?

Answer 3

Too slow

Question 4

How does the R-CNN ( Region based CNN) work?

Answer 4

1. Get ROI (Regions of interesest) from a sperate algorithm 2. Forward each region trough a convonet
2. Use SVM for classification and Bbox regression

Question 5

What is the probem for the R-CNN?

Answer 5

Training and inference is slow, training is ad-hoc

Question 6

What is the SPP methode for object detection?

Answer 6

1. Send the whole image trough a convonet.
2. Extract ROI from the result.
3. Use pooling to reduce the size of the ROI.
4. Use fully connected to svm for classification and fully connected to BB regression.

Question 7

What is the main advantage of SPP(Spatial pyramide pooling) over R-CNN?

Answer 7

Makes testing phase faster as we only need to use convo once. Training is still slow and ad-hoc, but faster than R-CNN

Question 8

What is the fast R-CNN algorithm?

Answer 8

1. Use convonet on image.
2. Do RoI proposals
3. Do RoI pooling
4. Use FC + linear for regression and FC + linear + softmax for classification

Question 9

What kind of loss does the fast R-CNN use?

Answer 9

cross entropy for classification and L1 for regression

Question 10

How does the Fast R-CNN RoI pooling work?

Answer 10

Divide the project proposal into 7x7 grid and do max pooling

Question 11

What is the main advantage of Fast R-CNN over SPP and R-CNN?

Answer 11

It’s fully trainable, with “fast” training and inference time.

Question 12

What is the main difference between Fast R-CNN and faster R-CNN?

Answer 12

Faster R-CNN trains a convo net to do proposal selection. The proposal convo net uses classification (object/not) and BB regression loss

Question 13

What is the mask R-CNN?

Answer 13

It adds FCN part to the faster R-Cnn to do semantic segmentation

Question 14

Describe the YOLO algorithm

Answer 14

1. Split image into grid (e.g. 7x7)
2. For each cell make two (or more) boundingboxes and predict p(object)
3. Each cell also predicts a class probability p(class | object)
4. Combine the bouding boxes and class prediction
5. Perform NMS

Question 15

How are the bounding boxes trained in Yolo?

Answer 15

For each cell find the best bounding box, adjust it and increase confidence. For cells without objects and other boxes reduce confidence.