R2D2: Repeatable and Reliable Detector and Descriptor

R2D2: Repeatable and Reliable Detector and Descriptor. Revaud, J., Weinzaepfel, P., De Souza, C., Pion, N., Csurka, G., Cabon, Y., & Humenberger, M. Advances in Neural Information Processing Systems, Neural information processing systems foundation, 6, 2019.

Paper

R2D2: Repeatable and Reliable Detector and Descriptor [link]

Website doi abstract bibtex

Interest point detection and local feature description are fundamental steps in many computer vision applications. Classical methods for these tasks are based on a detect-then-describe paradigm where separate handcrafted methods are used to first identify repeatable keypoints and then represent them with a local descriptor. Neural networks trained with metric learning losses have recently caught up with these techniques, focusing on learning repeatable saliency maps for keypoint detection and learning descriptors at the detected keypoint locations. In this work, we argue that salient regions are not necessarily discriminative, and therefore can harm the performance of the description. Furthermore, we claim that descriptors should be learned only in regions for which matching can be performed with high confidence. We thus propose to jointly learn keypoint detection and description together with a predictor of the local descriptor discriminativeness. This allows us to avoid ambiguous areas and leads to reliable keypoint detections and descriptions. Our detection-and-description approach, trained with self-supervision, can simultaneously output sparse, repeatable and reliable keypoints that outperforms state-of-the-art detectors and descriptors on the HPatches dataset. It also establishes a record on the recently released Aachen Day-Night localization dataset.

@article{
 title = {R2D2: Repeatable and Reliable Detector and Descriptor},
 type = {article},
 year = {2019},
 volume = {32},
 websites = {https://arxiv.org/abs/1906.06195v2},
 month = {6},
 publisher = {Neural information processing systems foundation},
 day = {14},
 id = {ab91b0a1-6228-3263-bc26-79a863cacf92},
 created = {2022-11-18T06:37:32.262Z},
 accessed = {2022-11-18},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2022-11-18T10:40:44.573Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {5a010301-acb6-4642-a6b2-8afaee1b741c,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
 private_publication = {false},
 abstract = {Interest point detection and local feature description are fundamental steps
in many computer vision applications. Classical methods for these tasks are
based on a detect-then-describe paradigm where separate handcrafted methods are
used to first identify repeatable keypoints and then represent them with a
local descriptor. Neural networks trained with metric learning losses have
recently caught up with these techniques, focusing on learning repeatable
saliency maps for keypoint detection and learning descriptors at the detected
keypoint locations. In this work, we argue that salient regions are not
necessarily discriminative, and therefore can harm the performance of the
description. Furthermore, we claim that descriptors should be learned only in
regions for which matching can be performed with high confidence. We thus
propose to jointly learn keypoint detection and description together with a
predictor of the local descriptor discriminativeness. This allows us to avoid
ambiguous areas and leads to reliable keypoint detections and descriptions. Our
detection-and-description approach, trained with self-supervision, can
simultaneously output sparse, repeatable and reliable keypoints that
outperforms state-of-the-art detectors and descriptors on the HPatches dataset.
It also establishes a record on the recently released Aachen Day-Night
localization dataset.},
 bibtype = {article},
 author = {Revaud, Jerome and Weinzaepfel, Philippe and De Souza, César and Pion, Noe and Csurka, Gabriela and Cabon, Yohann and Humenberger, Martin},
 doi = {10.48550/arxiv.1906.06195},
 journal = {Advances in Neural Information Processing Systems}
}

Downloads: 0

{"_id":"jr9EKmkoShKyQn2yL","bibbaseid":"revaud-weinzaepfel-desouza-pion-csurka-cabon-humenberger-r2d2repeatableandreliabledetectoranddescriptor-2019","author_short":["Revaud, J.","Weinzaepfel, P.","De Souza, C.","Pion, N.","Csurka, G.","Cabon, Y.","Humenberger, M."],"bibdata":{"title":"R2D2: Repeatable and Reliable Detector and Descriptor","type":"article","year":"2019","volume":"32","websites":"https://arxiv.org/abs/1906.06195v2","month":"6","publisher":"Neural information processing systems foundation","day":"14","id":"ab91b0a1-6228-3263-bc26-79a863cacf92","created":"2022-11-18T06:37:32.262Z","accessed":"2022-11-18","file_attached":"true","profile_id":"235249c2-3ed4-314a-b309-b1ea0330f5d9","group_id":"5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1","last_modified":"2022-11-18T10:40:44.573Z","read":"true","starred":false,"authored":false,"confirmed":false,"hidden":false,"folder_uuids":"5a010301-acb6-4642-a6b2-8afaee1b741c,bd3c6f2e-3514-47cf-bc42-12db8b9abe45","private_publication":false,"abstract":"Interest point detection and local feature description are fundamental steps\nin many computer vision applications. Classical methods for these tasks are\nbased on a detect-then-describe paradigm where separate handcrafted methods are\nused to first identify repeatable keypoints and then represent them with a\nlocal descriptor. Neural networks trained with metric learning losses have\nrecently caught up with these techniques, focusing on learning repeatable\nsaliency maps for keypoint detection and learning descriptors at the detected\nkeypoint locations. In this work, we argue that salient regions are not\nnecessarily discriminative, and therefore can harm the performance of the\ndescription. Furthermore, we claim that descriptors should be learned only in\nregions for which matching can be performed with high confidence. We thus\npropose to jointly learn keypoint detection and description together with a\npredictor of the local descriptor discriminativeness. This allows us to avoid\nambiguous areas and leads to reliable keypoint detections and descriptions. Our\ndetection-and-description approach, trained with self-supervision, can\nsimultaneously output sparse, repeatable and reliable keypoints that\noutperforms state-of-the-art detectors and descriptors on the HPatches dataset.\nIt also establishes a record on the recently released Aachen Day-Night\nlocalization dataset.","bibtype":"article","author":"Revaud, Jerome and Weinzaepfel, Philippe and De Souza, César and Pion, Noe and Csurka, Gabriela and Cabon, Yohann and Humenberger, Martin","doi":"10.48550/arxiv.1906.06195","journal":"Advances in Neural Information Processing Systems","bibtex":"@article{\n title = {R2D2: Repeatable and Reliable Detector and Descriptor},\n type = {article},\n year = {2019},\n volume = {32},\n websites = {https://arxiv.org/abs/1906.06195v2},\n month = {6},\n publisher = {Neural information processing systems foundation},\n day = {14},\n id = {ab91b0a1-6228-3263-bc26-79a863cacf92},\n created = {2022-11-18T06:37:32.262Z},\n accessed = {2022-11-18},\n file_attached = {true},\n profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},\n group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},\n last_modified = {2022-11-18T10:40:44.573Z},\n read = {true},\n starred = {false},\n authored = {false},\n confirmed = {false},\n hidden = {false},\n folder_uuids = {5a010301-acb6-4642-a6b2-8afaee1b741c,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},\n private_publication = {false},\n abstract = {Interest point detection and local feature description are fundamental steps\nin many computer vision applications. Classical methods for these tasks are\nbased on a detect-then-describe paradigm where separate handcrafted methods are\nused to first identify repeatable keypoints and then represent them with a\nlocal descriptor. Neural networks trained with metric learning losses have\nrecently caught up with these techniques, focusing on learning repeatable\nsaliency maps for keypoint detection and learning descriptors at the detected\nkeypoint locations. In this work, we argue that salient regions are not\nnecessarily discriminative, and therefore can harm the performance of the\ndescription. Furthermore, we claim that descriptors should be learned only in\nregions for which matching can be performed with high confidence. We thus\npropose to jointly learn keypoint detection and description together with a\npredictor of the local descriptor discriminativeness. This allows us to avoid\nambiguous areas and leads to reliable keypoint detections and descriptions. Our\ndetection-and-description approach, trained with self-supervision, can\nsimultaneously output sparse, repeatable and reliable keypoints that\noutperforms state-of-the-art detectors and descriptors on the HPatches dataset.\nIt also establishes a record on the recently released Aachen Day-Night\nlocalization dataset.},\n bibtype = {article},\n author = {Revaud, Jerome and Weinzaepfel, Philippe and De Souza, César and Pion, Noe and Csurka, Gabriela and Cabon, Yohann and Humenberger, Martin},\n doi = {10.48550/arxiv.1906.06195},\n journal = {Advances in Neural Information Processing Systems}\n}","author_short":["Revaud, J.","Weinzaepfel, P.","De Souza, C.","Pion, N.","Csurka, G.","Cabon, Y.","Humenberger, M."],"urls":{"Paper":"https://bibbase.org/service/mendeley/bfbbf840-4c42-3914-a463-19024f50b30c/file/40cff6ec-4f7b-281f-5ba3-4657f263c89e/full_text.pdf.pdf","Website":"https://arxiv.org/abs/1906.06195v2"},"biburl":"https://bibbase.org/service/mendeley/bfbbf840-4c42-3914-a463-19024f50b30c","bibbaseid":"revaud-weinzaepfel-desouza-pion-csurka-cabon-humenberger-r2d2repeatableandreliabledetectoranddescriptor-2019","role":"author","metadata":{"authorlinks":{}},"downloads":0},"bibtype":"article","biburl":"https://bibbase.org/service/mendeley/bfbbf840-4c42-3914-a463-19024f50b30c","dataSources":["2252seNhipfTmjEBQ"],"keywords":[],"search_terms":["r2d2","repeatable","reliable","detector","descriptor","revaud","weinzaepfel","de souza","pion","csurka","cabon","humenberger"],"title":"R2D2: Repeatable and Reliable Detector and Descriptor","year":2019}