DGD-net: Local Descriptor Guided Keypoint Detection Network

Abstract

In recent years, learning-based feature detection network has greatly improved the performance of keypoints matching. However, existing approaches have not fully utilized the representational ability of learned descriptors for feature detection. We propose a novel keypoint detection and description approach to make more use of the reliability of descriptors in matching on keypoint detection. We utilize the descriptor training loss to construct a guided score, which depicts the matching reliability of the descriptors, to supervise the detectors learning. We also propose a descriptor training loss to adapt to our detector training. In order to improve localization accuracy from the low-resolution feature map, we propose a method based on the idea of backtracking. Extensive experiments show the effectiveness of the proposed approach.

Self-supervised endoscopic image key-points matching

“…Recently, a dense feature descriptor namely DGD-net is designed in (Liu et al, 2021) where the training is guided by the reliability of the descriptor in matching. DGD-net proposes a backtracking method to enhance the localization accuracy.…”

ReDFeat: Recoupling Detection and Description for Multimodal Feature Learning

“…Secondly, 𝑎(R ( 𝒅 𝑖 ; 𝑫, 𝑫 )) only constrains the peaking of the detection, which would not suppress any detected probability of hard descriptors with large R ( 𝒅 𝑖 ; 𝑫, 𝑫 ) and solve the problems mentioned above [12], [9], [27], [34]. Thirdly, edge-based priori 𝑴 ( 𝑰) is introduced to balance the peaking constraint, instead of forcing the model to detect corners or edges [11], [20], [4]. Moreover, the weights are normalized by the expectation dynamically, so the weights would not be zeros and keep functioning.…”

“…Undoubtedly, those descriptors are limited by the detected scale, orientation and so on. To fill the gap between the detection and the description, the one-stage pipeline [11], [12], [27], [6], [34], [20], [9], [10], [36], [31] that learns to output dense detection scores and descriptors is proposed and further improvements are achieved.…”

“…Both the visible and the handcrafted multimodal features focus on corner or edge detection and description, which are believed to contain information that is invariant to geometric and radiant distortion. Their successes motivate many deep learning methods [11], [20], [4] and us to inject edge-based priori into the training.…”

GCNKD-Net:Multi-layer graph convolution networks for point cloud keypoints detection

“…Therefore, deep learning based 3D keypoint detectors are relatively rare compared to their growing success in learning 3D keypoint descriptors [21][22][23][24]. In addition, existing deep learning based methods generally use convolutional neural networks (CNNs) to extract the features for detecting keypoints in point clouds [20]. However, CNN is more suitable for 2D images with regular structures.…”