L2-Net: Deep Learning of Discriminative Patch Descriptor in Euclidean Space

By Yurun Tian, Bin Fan, Fuchao Wu. 2017

574 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Learning Spread-Out Local Feature Descriptors (2017, background): “Recently, there are a lot of works such as smart sampling strategies [1, 12] and structured loss [23] that improve the triplet loss.”
LF-Net: Learning Local Features from Images (2018, background): “However, many other approaches target some of their individual components, be it feature point extraction [31, 44], orientation estimation [50], or descriptor generation [36, 40, 41].”
Image Matching from Handcrafted to Deep Features: A Survey (2020, background): “In L2Net (Tian et al. 2017), Tian et al. applied a progressive sampling strategy to optimize the relative distance-based loss function in the Euclidean space.”
Learning fine-grained features via a CNN Tree for Large-scale Classification (2015, background): “Convolutional neural networks (CNN) [1, 2] have recently demonstrated superior performance on many tasks such as image classification [3, 4, 5], object detection [6, 7, 8, 9, 10, 11], object tracking [12, 13, 14], text detection [15, 16], text recognition [17, 18, 19], local feature description [20]”
SDC – Stacked Dilated Convolution: A Unified Descriptor Network for Dense Matching Tasks (2019, methodology): “• L2Net : The basic variant of [37] with only a single stream and without batch normalization, which we found to perform the best among all variants of this network.”
DISK: Learning local features with policy gradient (2020, methodology): “We consider several descriptors on DoG keypoints: RootSIFT [20, 2] L2-Net [39], HardNet [25], GeoDesc [22], SOSNet [40] and LogPolarDesc [13].”
The Perfect Match: 3D Point Cloud Matching With Smoothed Densities (2018, methodology): “In our implementation, we follow [36] and fix the affine parameters of the batch normalization layer to 1 and 0 and we do not train them during the training of the network.”
ASLFeat: Learning Local Features of Accurate Shape and Localization (2020, methodology): “To reduce computations, we replace the VGG backbone [34] used in D2-Net with more light-weight L2-Net [36].”
ASpanFormer: Detector-Free Image Matching with Adaptive Span Transformer (2022, methodology): “As is shown in previous works [10,8,9,12,28,27,26,29], CNN exhibits strong capability to capture local context and generates high-level features, which can be directly used to perform nearest neighbour matching.”
Uninformed Students: Student-Teacher Anomaly Detection With Discriminative Latent Embeddings (2019, methodology): “[35], we minimize the correlation between descriptors within one minibatch of inputs p to increase the descriptors’ compactness and remove unnecessary redundancy:”
Image Matching Across Wide Baselines: From Paper to Practice (2020, background): “ [55], MSER [53]; and their afﬁne-covariant versions: DoGAfﬁne, Hessian-Afﬁne [55,14], DoG-AffNet [59], and Hessian-AffNet [59]. The second group includes descriptors learned on DoG keypoints: L2-Net [89], Hardnet [57], Geodesc [50], SOSNet [90], ContextDesc [49], and LogPolarDesc [34]. The last gro”
SOSNet: Second Order Similarity Regularization for Local Descriptor Learning (2019, methodology): “, DeepDesc(DDesc) [34], TFeat [2], L2Net [36], HardNet(HNet) [26], HardNet with GOR [41], Scale-Aware Descriptor [17], DOAP [15] and GeoDesc [22].”
R2D2: Repeatable and Reliable Detector and Descriptor (2019, methodology): “The used L2-Net architecture [55] is applied patch by patch, which is quite slow.”
Working hard to know your neighbor's margins: Local descriptor learning loss (2017, methodology): “We were able to get reasonable results with random and hard negative mining sampling only with additional correlation penalty on descriptor channels (CPR), as proposed in [24].”