CornerNet is a novel object detection method that detects objects as pairs of corners—top-left and bottom-right corners of bounding boxes—using a single convolutional neural network (CNN). This approach eliminates the need for anchor boxes commonly used in prior single-stage detectors. Instead, CornerNet uses corner pooling, a new type of pooling layer that helps the network better localize corners. The network predicts heatmaps for top-left and bottom-right corners, along with embeddings for each detected corner, which are used to group corners belonging to the same object. Experiments show that CornerNet achieves a 42.2% AP on MS COCO, outperforming all existing one-stage detectors. CornerNet is inspired by associative embedding methods and introduces corner pooling to enhance corner detection. The method is efficient and effective, with ablation studies showing that corner pooling is critical to its performance. The network is trained using a modified focal loss and uses an hourglass network as its backbone. CornerNet is evaluated on the MS COCO dataset and demonstrates competitive results compared to two-stage detectors. The approach simplifies the output of the network and eliminates the need for designing anchor boxes. The method is effective in detecting corners across different image quadrants and improves performance for medium and large objects. CornerNet achieves high-quality bounding boxes with tight coverage, outperforming other state-of-the-art detectors in terms of AP at high IoU thresholds. The main bottleneck for CornerNet is detecting corners accurately.CornerNet is a novel object detection method that detects objects as pairs of corners—top-left and bottom-right corners of bounding boxes—using a single convolutional neural network (CNN). This approach eliminates the need for anchor boxes commonly used in prior single-stage detectors. Instead, CornerNet uses corner pooling, a new type of pooling layer that helps the network better localize corners. The network predicts heatmaps for top-left and bottom-right corners, along with embeddings for each detected corner, which are used to group corners belonging to the same object. Experiments show that CornerNet achieves a 42.2% AP on MS COCO, outperforming all existing one-stage detectors. CornerNet is inspired by associative embedding methods and introduces corner pooling to enhance corner detection. The method is efficient and effective, with ablation studies showing that corner pooling is critical to its performance. The network is trained using a modified focal loss and uses an hourglass network as its backbone. CornerNet is evaluated on the MS COCO dataset and demonstrates competitive results compared to two-stage detectors. The approach simplifies the output of the network and eliminates the need for designing anchor boxes. The method is effective in detecting corners across different image quadrants and improves performance for medium and large objects. CornerNet achieves high-quality bounding boxes with tight coverage, outperforming other state-of-the-art detectors in terms of AP at high IoU thresholds. The main bottleneck for CornerNet is detecting corners accurately.