Comparative Evaluation of Hand-Crafted and Learned Local Features

By Johannes L. Schönberger, Hans Hardmeier, Torsten Sattler, Marc Pollefeys. Computer Vision and Pattern Recognition, 2017

314 citations recorded by the source corpus.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/911e91af256787bfe9d703c75262c4ebc8d8c6d6

How researchers see this paper

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

In Aerospace Engineering

In Computer Vision and Pattern Recognition

Citations in the corpus

Learning to Find Good Correspondences (2017, result): “form earlier ones on point-matching benchmarks, incorporating them into pose estimation pipelines may not necessarily translate into a performance increase, as indicated by two recent studies [25, 3].”
Grounding Image Matching in 3D with MASt3R (2024, background): “Overall, keypoint-based approaches are predominant in many benchmarks [7,35,44,77], underscoring their enduring value in tasks requiring high precision and speed [19,77].”
Neighbourhood Consensus Networks (2018, unknown): “However, unlike in other visual recognition tasks, such as image classiﬁcation or object detection, where trainable image representations have become the de facto standard, the performance gains obtained by trainable features over the classic hand-crafted ones have been only modest at best [35].”
Kornia: an Open Source Differentiable Computer Vision Library for PyTorch (2019, background): “SIFT descriptor is still one of the best for 3d reconstruction [50] tasks.”
ASLFeat: Learning Local Features of Accurate Shape and Localization (2020, methodology): “Finally, we extensively evaluate our methods across various practical scenarios, including image matching [1, 2], 3D reconstruction [32] and visual localization [30].”
LF-Net: Learning Local Features from Images (2018, background): “One problem with this approach is that increasing the performance of one component does not necessarily translate into overall improvements [35, 48].”
Visual Camera Re-Localization From RGB and RGB-D Images Using DSAC (2020, background): “Learning-based sparse feature pipelines [41], [42], [43], [44] might ultimately be able to overcome these issues, but currently it is an open research question whether learned sparse features consistently exceed the capabilities of their hand-crafted predecessors [45], [46].”
Working hard to know your neighbor's margins: Local descriptor learning loss (2017, background): “Recent studies have confirmed that SIFT and its variants (RootSIFT-PCA [16], DSP-SIFT [17]) significantly outperform learned descriptors in image matching and small-scale retrieval [18], as well as in 3D-reconstruction [19].”
Image Matching Across Wide Baselines: From Paper to Practice (2020, background): “A large-scale benchmark for SfM was proposed in [80] – without ground truth.”
SOSNet: Second Order Similarity Regularization for Local Descriptor Learning (2019, methodology): “Similarly to the observations in [32, 22], SIFT achieves the smallest reprojection error on all tests, thus demonstrating that it is still an attractive choice for image matching.”
Image Matching from Handcrafted to Deep Features: A Survey (2020, methodology): “Descriptorswith deep learning techniques can be regarded as an extension of those based on classical learning (Schonberger et al. 2017).”
A review of multimodal image matching: Methods and applications (2021, background): “The primary goal is to learn a representation that can enable the two matched features to be as close as possible while the unmatched ones are far apart in the measuring space [219].”
From Coarse to Fine: Robust Hierarchical Localization at Large Scale (2018, background): “We also report the track length, i.e. the number of observation per 3D point, as deﬁned by [48].”
DISK: Learning local features with policy gradient (2020, background): “While not immune to the deep learning “revolution”, 3D reconstruction is one of the last bastions where sparse, hand-crafted solutions remain competitive with or outperform their dense, learned counterparts [37, 34, 16].”