Object Detection

1. Two rectangles R1 and R2 have left-up and right-down points A and B, C and D accordingly. Coordinates of points: A (0, 77), B (23, 26), C (15, 51), D (41, 0). Compute IoU of these rectangles in percents. Round answer to the nearest integer in percents.

Consider face detector have Miss rate 0.40 for FPPI = 10^-1. We are working with dataset that has 5 faces on each image in average. What Precision and Recall corresponds to this parameters? Enter precision and recall values in percentages with space:

Consider constructing pyramid for sliding window object detection. We’d like to use window with size 20×20 pixels and find objects with size from 20 to 50 pixels. Images in pyramid are upscaled with factor 1.1 (i.e. by 10% each time). How many images (including original, not scaled, image) are needed for this pyramid?

What data augmentation methods are useful for training HOG+SVM or VJ face detector?

We work with video of size 1024×768 pixels and 25 fps. We use sliding window object detector with window size 20×20 pixels and stride 2, 1 image scale (i.e. without pyramid). What should be false positive rate of the detector s.t. detector output false positive less frequently than 1 time per second? Round answer with 1e-07 precision.

Select correct sentences for R-CNN object detection architecture:

1. Select correct sentences for Faster R-CNN object detection architecture:

How many numbers will YOLO detector output per image if the model has 6 x 10 grid, every cell has 3 position hypotheses and there are 25 object classes in the training sample?