Edge Boxes is a novel method for generating object bounding box proposals using edges. The method leverages the sparse yet informative nature of edges to identify potential object locations. The key observation is that the number of contours fully enclosed by a bounding box indicates the likelihood of the box containing an object. A simple box objectness score is proposed, which measures the number of edges inside the box minus those that are part of contours overlapping the box's boundary. This score is computed efficiently using data structures, allowing for the rapid evaluation of millions of candidate boxes, returning a ranked set of top-scoring proposals. The method achieves high accuracy and efficiency, with results showing over 96% object recall at an overlap threshold of 0.5 and over 75% recall at 0.7 using just 1000 proposals. The approach runs in 0.25 seconds and a near real-time variant runs in 0.1 seconds with minimal accuracy loss. The method uses a structured edge detector to generate edge maps, which are then clustered into edge groups. Affinities between edge groups are computed based on their positions and orientations. The score for a box is calculated by summing the edge strength of groups within the box and subtracting the strength of groups that overlap the box's boundary. A sliding window approach is used to generate candidate boxes, which are then refined using a coarse-to-fine search. The method outperforms state-of-the-art approaches in terms of accuracy and efficiency, achieving higher recall and faster processing times. Results show that Edge Boxes performs well across various IoU thresholds and numbers of object proposals, with the Edge Boxes 70 variant performing particularly well. The method is efficient, with a runtime of 0.25 seconds for the baseline approach and 0.09 seconds for the near real-time variant. The approach is also effective in generating object proposals for a wide range of applications, including object detection and segmentation.Edge Boxes is a novel method for generating object bounding box proposals using edges. The method leverages the sparse yet informative nature of edges to identify potential object locations. The key observation is that the number of contours fully enclosed by a bounding box indicates the likelihood of the box containing an object. A simple box objectness score is proposed, which measures the number of edges inside the box minus those that are part of contours overlapping the box's boundary. This score is computed efficiently using data structures, allowing for the rapid evaluation of millions of candidate boxes, returning a ranked set of top-scoring proposals. The method achieves high accuracy and efficiency, with results showing over 96% object recall at an overlap threshold of 0.5 and over 75% recall at 0.7 using just 1000 proposals. The approach runs in 0.25 seconds and a near real-time variant runs in 0.1 seconds with minimal accuracy loss. The method uses a structured edge detector to generate edge maps, which are then clustered into edge groups. Affinities between edge groups are computed based on their positions and orientations. The score for a box is calculated by summing the edge strength of groups within the box and subtracting the strength of groups that overlap the box's boundary. A sliding window approach is used to generate candidate boxes, which are then refined using a coarse-to-fine search. The method outperforms state-of-the-art approaches in terms of accuracy and efficiency, achieving higher recall and faster processing times. Results show that Edge Boxes performs well across various IoU thresholds and numbers of object proposals, with the Edge Boxes 70 variant performing particularly well. The method is efficient, with a runtime of 0.25 seconds for the baseline approach and 0.09 seconds for the near real-time variant. The approach is also effective in generating object proposals for a wide range of applications, including object detection and segmentation.