Skip to content
ml-recall

#offsets

2 questions from Object Detection

Timed test

2 questions · 2 min · score and explanations at the end

Take a test
Q 1 / 2Score 0 / 0

Question 1: Short answer

mediumAI-checkedObject Detection

d2l labels the offset of an anchor AA (centre (xa,ya)(x_a, y_a), size wa×haw_a \times h_a) towards its ground-truth box BB (centre (xb,yb)(x_b, y_b), size wb×hbw_b \times h_b) as ((xb−xa)/waσx,(yb−ya)/haσy,log⁡(wb/wa)σw,log⁡(hb/ha)σh)\left(\frac{(x_b - x_a)/w_a}{\sigma_x}, \frac{(y_b - y_a)/h_a}{\sigma_y}, \frac{\log(w_b/w_a)}{\sigma_w}, \frac{\log(h_b/h_a)}{\sigma_h}\right) with σx=σy=0.1\sigma_x = \sigma_y = 0.1 and σw=σh=0.2\sigma_w = \sigma_h = 0.2. The anchor has centre (100,100)(100, 100) and size 50×10050 \times 100; the ground truth has centre (110,90)(110, 90) and size 60×8060 \times 80 (width × height).

The xx offset is and the width offset is .

Fill in every blank (hints in the boxes). Enter moves to the next blank.