Question 1 of 20

Semantic segmentation assigns:

Question 2 of 20

U-Net skip connections copy encoder features to the decoder to:

Question 3 of 20

Instance segmentation differs from semantic segmentation by:

Question 4 of 20

Dice loss is popular for segmentation when:

Question 5 of 20

Mean IoU (mIoU) for segmentation averages:

Question 6 of 20

Encoder–decoder architecture in segmentation:

Question 7 of 20

ROI Align in Mask R-CNN improves over ROI Pool by:

Question 8 of 20

Focal loss modifies cross-entropy to:

Question 9 of 20

Panoptic segmentation combines:

Question 10 of 20

Boundary F-score in segmentation evaluation:

Question 11 of 20

Transposed convolution (deconv) in decoders:

Question 12 of 20

Mask R-CNN adds to Faster R-CNN:

Question 13 of 20

Per-pixel cross-entropy segmentation loss with class weights:

Question 14 of 20

DeepLab atrous (dilated) convolutions enlarge receptive field by:

Question 15 of 20

Ignoring void/unlabeled pixels in mIoU means:

Question 16 of 20

Compared to classification, segmentation training labels are:

Question 17 of 20

Mask IoU matching in instance evaluation requires:

Question 18 of 20

Combining CE and Dice loss:

Question 19 of 20

SegFormer and transformer segmentation encoders use:

Question 20 of 20

Qualitative segmentation review should include: