They can be predicted without problems.

They can be predicted if and they belong to the same class.

They can be predicted if they belong to the same class.

They can be predicted if .

The -th bounding-box predictor in cell is responsible for the ground-truth object prediction.

An object of class is present in cell .

The -th bounding-box predictor in cell is responsible for an object prediction.

Cell contains exactly objects.

R-CNN achieved the highest average precision on both artwork datasets.

YOLO achieved higher average precision than DPM and R-CNN on both artwork datasets, supporting stronger cross-domain generalization in this experiment.

All compared methods lost the same proportion of their performance.

YOLO achieved good results only after being fine-tuned on each artwork dataset.

Each bounding-box predictor predicts one confidence score, but only the predictor with the highest score is used.

Each bounding-box predictor predicts a set of conditional class probabilities, but only the class with the highest probability is selected.

Each cell predicts one set of conditional class probabilities, shared by its bounding-box predictors.

Each cell predicts only one confidence score in total, shared by its bounding-box predictors.

increases the coordinate-loss weight, while decreases the confidence-loss weight for boxes without objects.

decreases the confidence-loss contribution from boxes without objects.

Taking the square roots of and makes a fixed absolute size error more costly for a large box than for a small box.

increases the weight of the classification error.

All four values are absolute pixel coordinates in the original image.

The center coordinates are relative to the bounds of the grid cell, while are relative to the whole image.

All four values are relative to the bounds of the grid cell.

The center coordinates are relative to the whole image, while are relative to the grid cell.

It sees the entire image during both training and inference.

It uses Selective Search to inspect the most relevant regions.

All answers are wrong.

It classifies every image patch independently and then combines them.

YOLO requires batch processing to achieve real-time speed.

Both models run below 30 frames per second.

Fast YOLO is slower than Base YOLO but more accurate.

Base YOLO runs at about 45 frames per second, while Fast YOLO runs at more than 150 frames per second.

The two models' bounding-box coordinates are averaged for every detection.

YOLO is retrained using Fast R-CNN’s region proposals.

If YOLO predicts a similar box, the Fast R-CNN detection is boosted using YOLO's predicted probability and the overlap between the boxes.

The Fast R-CNN detection is deleted whenever YOLO predicts a similar box.