Expert Sample Consensus Applied to Camera Re-Localization

Carsten Rother; Eric Brachmann

REVIEW 2 major objections 4 minor 54 references

Expert Sample Consensus Applied to Camera Re-Localization

T0 review · 2 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A mixture-of-experts pose estimator that splits its hypothesis budget by gating and picks the winner by sample consensus reaches 88.1% re-localization on 19 combined rooms.

desk verdict A coherent and well-motivated ensembling method for DSAC, with solid experiments; the main caveat is an unmeasured sensitivity to gating misses. read the letter →

arxiv 1908.02484 v1 pith:GZOUBGR3 submitted 2019-08-07 cs.CV

Eric Brachmann , Carsten Rother This is my paper · ORCID

classification cs.CV

keywords ExpertSampleConsensusMixtureofExpertscamerare-localizationdifferentiableRANSACscenecoordinateregressionend-to-endtraining6Dposeestimation

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that, in a mixture-of-experts system for fitting parametric models, the gating network should neither choose a single expert nor average expert outputs. Instead, it should convert gating probabilities into a multinomial split of a fixed hypothesis budget, and the final estimate should be selected by sample consensus across all experts' hypotheses. This makes the ensemble forgiving of gating mistakes while retaining specialization, and it can be trained jointly and end-to-end. Applied to camera re-localization, the method reaches 88.1% correctly re-localized images on the combined 19Scenes environment, against 53.3% for the single-network baseline and 47.5% for hard expert selection.

What carries the argument

The central mechanism is the multinomial hypothesis split combined with a cross-expert softmax selection. Given M expert networks, each predicting scene coordinates, and a gating distribution g(e, I; w), ESAC draws a split H = (n_1, ..., n_M) with sum n_e = N from a multinomial distribution, gives expert e a budget of n_e pose hypotheses, and then selects the final pose by softmax over the soft inlier scores s(h_{e,j}, Y_e) of all N hypotheses. This makes the gating network's role one of resource allocation: as long as the correct expert receives a nonzero share, a good hypothesis can still win on geometric consistency. Training minimizes the expected task loss over both the split and the selection, with gradients approximated by sampling, including the log-derivative term for the split probability.

What would settle it

On a held-out subset of 19Scenes, force the gating probabilities so that the true scene's expert receives zero hypotheses on images the gating network currently gets wrong. If ESAC's accuracy does not fall below the reported 88.1% by a substantial margin, the paper's explanation that consensus rescues gating errors is not what is driving the result; if instead the accuracy collapses, the claim is supported.

Watch

Extended reading notes

Core claim

Expert Sample Consensus (ESAC) claims that the correct way to combine a mixture of experts with differentiable RANSAC is not to let the gating network choose a single expert, nor to average expert outputs, but to convert gating probabilities into a multinomial split of a fixed hypothesis budget and then to select the final pose by sample consensus across all experts' hypotheses. The gating network therefore only decides how many of the N pose hypotheses each expert may propose; the geometric consistency score decides which hypothesis wins. On the combined 19-room indoor benchmark (7Scenes plus 12Scenes), ESAC re-localizes 88.1% of test images within 5 cm and 5 degrees, compared with 53.3% for the single-network DSAC++ baseline and 47.5% for hard expert selection, and it nearly matches an oracle that is told the correct scene (89.0%). The authors also report large gains over DSAC++ on outdoor datasets Dubrovnik and Aachen Day, while noting that sparse feature-based localization remains more accurate there.

Load-bearing premise

The load-bearing premise is that the gating network gives the correct expert at least some of the hypothesis budget on every test image, and that the soft inlier score is a fair, comparable measure of hypothesis quality across different experts.

Editorial extensions

If this is right

On the combined 19-room environment, distributing hypotheses by gating reaches 88.1% correct re-localization versus 53.3% for a single network and 47.5% for hard expert selection, with the gap growing as the environment becomes larger.
Training the ensemble end-to-end with the ESAC objective matters most for precision: at a stricter 2 cm / 2 degree threshold, joint end-to-end training improves accuracy by more than 10 percentage points over the initialization.
Because experts with zero assigned hypotheses do not need to run, ESAC is faster than the monolithic baseline (555 ms versus 940 ms per image on 19Scenes) and can cap the number of active experts, gaining 19.7 percentage points over hard selection when only the top two experts are used.
The same formulation transfers to large outdoor scenes by dividing the environment into clusters and training an expert per cluster; on the Aachen Day benchmark ESAC raises correct localization from 0.4% to 42.6% at the strictest threshold as the number of experts grows, while still trailing sparse feature-based methods.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The same budget-splitting idea should apply to other robust-fitting tasks with a discrete gate, such as object pose estimation or fundamental-matrix estimation, where the gating signal is weak or ambiguous.
Since only a nonzero share to the correct expert is needed, the gating network can be trained with substantially weaker supervision than a hard classifier; one testable consequence is that ESAC should tolerate noisier scene labels than expert selection.
The outdoor results suggest the remaining gap to feature-based localization is driven less by the ensemble design and more by the quality of the structure-from-motion ground truth used for training; filtering outlier 3D points before training is a natural follow-up.
The toy experiment's high model-type classification accuracy hints that ESAC could serve as a differentiable model-selection procedure, choosing not only parameters but the family of model, by letting experts compete through sample consensus.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague,

This paper does something genuinely new: it merges DSAC with mixture-of-experts by distributing the N pose hypotheses across experts according to a gating distribution, then selecting via sample consensus. The joint training objective (Eq. 12) and the gradient estimator (Eq. 13) are clearly derived, and the experiments on 19Scenes show a large margin over DSAC++ and naive expert selection. That is real. The toy problem and the confusion matrices support the idea that ESAC hedges against gating mistakes better than hard selection.

The main soft spot is exactly what the stress-test note flags: the robustness claim pivots on the gating network assigning nonzero probability to the correct expert, but the paper never reports the distribution of gating probabilities for the correct expert, top-k recall, or the fraction of failures where zero hypotheses went to the right expert. The gap between Oracle Gating (89.0%) and ESAC (88.1%) suggests the consensus selection is doing most of the work once the right expert gets some hypotheses, but without that decomposition the sensitivity of the method to gating misses is under-measured. That said, the limited max-expert results in Fig. 8 are indirect evidence that the correct expert often sits near the top of the gating ranking, so the concern is real but not damning.

Other soft spots are minor: the experiments appear to be single runs without variance, code is promised but not shipped, and the outdoor results, while improved, still trail classical feature-based methods. The authors candidly attribute this to SfM ground-truth outliers, which is a fair caveat.

On balance, this is a serious, well-written paper. The math is coherent, the comparisons are not circular, and the empirical claims are strong enough to justify a careful referee process. It deserves peer review, not desk rejection.

Referee Report

2 major / 4 minor

Summary. The paper proposes Expert Sample Consensus (ESAC), which combines a Mixture-of-Experts architecture with Differentiable Sample Consensus (DSAC) for robust model fitting. Rather than letting a gating network select a single expert or averaging expert outputs, ESAC samples a multinomial split of a fixed hypothesis budget across experts and then selects the final hypothesis by sample consensus across all experts. The full ensemble, including the gating network, is trained end-to-end by minimizing the expected task loss with a K=1 gradient estimator. The method is evaluated on a synthetic line/circle toy problem, on indoor camera re-localization using 7Scenes, 12Scenes and their union 19Scenes, and on outdoor Dubrovnik and Aachen. On 19Scenes, ESAC re-localizes 88.1% of test images compared to 53.3% for DSAC++ and 47.5% for hard expert selection. A limitation paragraph in the appendix discusses remaining gaps to classical feature-based localization outdoors.

Significance. The paper makes a simple but effective conceptual contribution: distributing hypotheses according to gating probabilities while performing final selection by geometric consensus. This is a natural synthesis of MoE and DSAC, and the experiments provide strong evidence for its practical value. The oracle-gating and uniform-gating ablations in Fig. 8 cleanly bracket the performance of the gating network, and the end-to-end training ablation in Fig. 7 shows the benefit of joint optimization. The authors also honestly discuss the limitations of scene-coordinate regression on SfM-based outdoor ground truth. The main threat to the paper's conclusions is empirical: all reported numbers are single runs without error bars, and the claimed robustness to gating errors is not directly measured.

major comments (2)

[5.2, Fig. 6 and Fig. 8] All accuracy numbers in the main comparison (Fig. 6) and in the speed/accuracy trade-off (Fig. 8) appear to be single-run point estimates; no error bars, confidence intervals, or significance tests are reported. This is particularly relevant for the toy experiment in Fig. 4, where the improvement of ESAC over Expert Selection is only +3.9%, and for the 19Scenes comparison where the margin is large but still subject to training stochasticity. Please report mean and standard deviation over multiple training runs, or at least a paired significance test, for the key comparisons (ESAC vs DSAC++ vs Expert Selection on 19Scenes, and ESAC vs Expert Selection on the toy problem).
[3.3, Eqs. (10)-(11); 5.2, Fig. 8] The paper's central robustness argument is that 'as long as the gating network attributes some probability to the correct expert' (Introduction), ESAC can recover from gating errors by giving that expert a nonzero share of the N hypotheses and selecting by consensus. This premise is never directly measured: the paper does not report the distribution of gating probabilities assigned to the ground-truth expert, the top-k recall of the gating network, or the frequency with which the correct expert receives zero hypotheses. For N=256, an expert with gating probability p=0.01 is excluded from the hypothesis pool with probability 7.6%, and with p=0.005 with probability 27.7%, so the operating regime matters. The comparison with Oracle Gating (89.0% vs 88.1%) shows that the total penalty for not knowing the scene ID is small, but it does not decompose that penalty into gating misses versus consensus-score errors when the correct expert did receive hypotheses. Adding this failure analysis would directly validate the claimed mechanism.

minor comments (4)

[5.2] The sentence 'It looses 3.5% accuracy' contains a typo: 'looses' should be 'loses'.
[Fig. 6b] Please define how 'Classification Accuracy' is computed for ESAC in Fig. 6b; it appears to be the scene label of the expert that produced the selected hypothesis, but the text should state this explicitly.
[Appendix B, Eq. (16)] In Eq. (16), the similarity measure has a slightly confusing arrangement of the exponent; please check that the formula as printed matches the intended Gaussian (e.g., whether sigma_e should be squared).
[Appendix B] The phrase 'we me minimize the KL-divergence' contains a typo; it should read 'we minimize'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: ESAC's hypothesis allocation and consensus selection are trained by an independent expected-loss objective, and self-citations to DSAC/DSAC++ serve as prior, independently evaluated building blocks.

full rationale

The paper's derivation chain is self-contained and empirically grounded. ESAC defines a gating distribution p(e), allocates a fixed budget of N hypotheses via the multinomial distribution in Eq. 10, selects the final pose by a softmax over sample-consensus scores across all experts in Eq. 11, and trains the ensemble by minimizing the expected task loss in Eq. 12. No equation reduces the target result to an input: the reported 88.1% accuracy on 19Scenes is a measured held-out result, not a fitted parameter renamed as a prediction. The gating network and experts are pretrained with scene labels and scene-coordinate supervision, but the ESAC objective then fine-tunes them end-to-end via gradients of Eq. 12, so the final accuracy is not forced by the initialization. Self-citations to DSAC [6] and DSAC++ [8] provide the differentiable RANSAC machinery and network design; these are prior, externally benchmarked building blocks rather than unverified premises that import the conclusion. The Oracle Gating baseline (89.0% vs. 88.1%) is an upper-bound comparison, not a circular construction. The skeptic's concern that gating top-k recall and zero-allocation probabilities are unreported is a legitimate empirical robustness or falsifiability limitation, but it is not a circularity: the paper's equations do not define the result in terms of itself. Accordingly, no specific circular step can be exhibited, and the appropriate finding is no significant circularity.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim rests on standard learning assumptions about specialization, plus two unproven practical assumptions: the K=1 gradient estimator and comparability of inlier scores across experts. All hyperparameters are inherited from DSAC++ or chosen by hand; none are fitted to the test set. No new physical or conceptual entities are introduced.

free parameters (6)

inlier threshold tau = 10 px
Controls the soft inlier count in Eq. 5; inherited from DSAC++ and used for all relocalization experiments, so the accuracy numbers depend on this choice.
task loss weight gamma = 100 (degrees vs meters)
Balances rotation and translation errors in the relocalization loss; taken from PoseNet and DSAC++, not tuned by the authors in this paper.
soft score hyperparameters alpha, beta = inherited from DSAC++ [8]
Control the softmax temperature of the inlier score in Eq. 5; the paper does not report specific values or sensitivity analysis.
number of experts M = 7, 12, 19 indoor; 10, 20, 50 outdoor
Chosen equal to the number of rooms or kMeans clusters; the central scalability comparison depends on these choices.
cluster softness gamma (outdoor) = 5
Controls the soft assignment of training images to kMeans clusters in Eq. 16 during outdoor expert initialization.
hypothesis budget N = 64 (toy), 256 (relocalization)
Number of model hypotheses sampled per image; follows DSAC conventions and affects the robustness and computation trade-off.

assumptions (4)

domain assumption The gating distribution p(e) reflects true scene identity well enough that the correct expert receives a nonzero share of hypotheses.
Used in Eq. 10 to sample hypothesis split H; if p(e)=0 for the correct expert, no correct hypotheses exist and sample-consensus selection cannot recover. Pretraining with scene labels is designed to ensure this.
domain assumption The soft inlier score s(h,Y) is comparable across different experts' hypotheses.
Eq. 11 takes a single softmax over all hypotheses from all experts; if scores have different scales per expert, selection is biased. The paper provides no analysis of score comparability.
ad hoc to paper A single Monte Carlo sample (K=1) of the gating split gives useful gradients for joint training.
Eq. 13 approximates the gradient expectation with K=1; the paper says this suffices in practice but provides no convergence or variance analysis.
domain assumption SfM ground truth for outdoor training images is accurate enough for dense scene coordinate regression.
Appendix B shows substantial outlier points in Dubrovnik and Aachen reconstructions and attributes ESAC's outdoor limitation partly to this noise; the central outdoor comparison depends on this assumption.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Expert Sample Consensus Applied to Camera Re-Localization." pith.science (2026). https://pith.science/paper/GZOUBGR3

@misc{pith2026190802484,
  author       = {Pith},
  title        = {Pith review of: Expert Sample Consensus Applied to Camera Re-Localization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GZOUBGR3}},
  note         = {Machine review of arXiv:1908.02484}
}

read the original abstract

Fitting model parameters to a set of noisy data points is a common problem in computer vision. In this work, we fit the 6D camera pose to a set of noisy correspondences between the 2D input image and a known 3D environment. We estimate these correspondences from the image using a neural network. Since the correspondences often contain outliers, we utilize a robust estimator such as Random Sample Consensus (RANSAC) or Differentiable RANSAC (DSAC) to fit the pose parameters. When the problem domain, e.g. the space of all 2D-3D correspondences, is large or ambiguous, a single network does not cover the domain well. Mixture of Experts (MoE) is a popular strategy to divide a problem domain among an ensemble of specialized networks, so called experts, where a gating network decides which expert is responsible for a given input. In this work, we introduce Expert Sample Consensus (ESAC), which integrates DSAC in a MoE. Our main technical contribution is an efficient method to train ESAC jointly and end-to-end. We demonstrate experimentally that ESAC handles two real-world problems better than competing methods, i.e. scalability and ambiguity. We apply ESAC to fitting simple geometric models to synthetic images, and to camera re-localization for difficult, real datasets.

Figures

Figures reproduced from arXiv: 1908.02484 by the authors.

**Figure 1.** Camera Re-Localization Using ESAC. Given an environment consisting of several ambiguous rooms (top) and a query image (middle), we estimate the 6D camera pose (bottom). A gating network (black) predicts a probability for each room. We distribute a budget of pose hypotheses to expert networks specialized to each room. We choose the pose hypothesis with maximum sample consensus (green), i.e. the maximum geometric … view at source ↗

**Figure 2.** Network Ensemble for a Toy Problem. a) Two expert networks, one specialized to finding lines, one specialized to finding circles. Both experts predict a set of 2D points which should lie on the line or circle, respectively. We fit model parameters to these points using differentiable RANSAC. b) The gating network predicts whether an image contains either a line or a circle. However, it is likely to make some mistak… view at source ↗

**Figure 3.** Ensemble Interplay. Given an image of a line or a circle, we estimate the parameters of the associated model. a) The gating network chooses one expert for a given input. We sample model hypotheses only based on this expert’s prediction. b) The gating network predicts how the number of model hypotheses should be divided among experts, i.e. we sample line and circle hypotheses. In this example, the estimate of a) and … view at source ↗

Figures from the paper (7 more)

**Figure 5.** Figure 5: Pose Accuracy when Scene ID is known. Percentage of pose estimates with an error below 5cm and 5◦ , and median errors. type. Here, ESAC outperforms Expert Selection by 11.5%. The good classification accuracy indicates that ESAC might be a suitable method for model sele…

**Figure 7.** Figure 7: Effect of End-to-End Training. Average re-localization accuracy of ESAC for 19Scenes when we train the entire ensemble of networks or parts of it end-to-end. Top: Acceptance threshold of 5cm and 5 ◦ . Bottom: 2cm and 2◦ . is significant but not large using the common a…

**Figure 8.** Figure 8: Accuracy vs. Speed on 19Scenes. We measure the average processing time for an image on a single Tesla K80 GPU including reading data. For ESAC, we can limit the maximum number of top ranked experts evaluated for a test image. Method Dubrovnik [27] Median Accuracy Aach…

**Figure 10.** Figure 10: Qualitative Result on Aachen. a) Clustering of the Aachen dataset used to initialize ESAC with M = 10 experts. b) Positions of hypotheses drawn by DSAC++ for the query image shown in a). Hypothesis sample consensus (hypothesis score) is indicated by color, normalized …

**Figure 11.** Figure 11: Scene Confusion. We compare confusion matrices of Expert Selection and ESAC for 7Scenes, 12Scenes and 19Scenes. The Y-axis shows the true scene, the X-axis shows the estimated scene. The scene ordering on the X-axis (from left to right) follows the Y-axis (from top to…

**Figure 12.** Figure 12: Ambiguous Scenes. We show test frames of scenes that Expert Selection confuses often (cf [PITH_FULL_IMAGE:figures/full_fig_p010_12.png]

**Figure 13.** Figure 13: Ground Truth SfM Reconstructions. Both datasets contain outlier 3D points in the ground truth reconstruction. The outlier ratio is substantial for Dubrovnik, and still noticeable for Aachen. [4] Relja Arandjelovic, Petr Gronat, Akihiko Torii, Tomas Pa- ´ jdla, and Jos…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 49 canonical work pages

[1]

Network of experts for large-scale image categoriza- tion

Karim Ahmed, Mohammad Haris Baig, and Lorenzo Torre- sani. Network of experts for large-scale image categoriza- tion. In ECCV, 2016. 2

work page 2016
[2]

A new look at the statistical model identi- ﬁcation

Hirotugu Akaike. A new look at the statistical model identi- ﬁcation. TAC, 1974. 3

work page 1974
[3]

Expert gate: Lifelong learning with a network of experts

Rahaf Aljundi, Punarjay Chakravarty, and Tinne Tuytelaars. Expert gate: Lifelong learning with a network of experts. In CVPR, 2017. 2, 3 Dubrovnik Aachen Figure 13. Ground Truth SfM Reconstructions. Both datasets contain outlier 3D points in the ground truth reconstruction. The outlier ratio is substantial for Dubrovnik, and still noticeable for Aachen

work page 2017
[4]

NetVLAD: CNN architecture for weakly supervised place recognition

Relja Arandjelovi ´c, Petr Gronat, Akihiko Torii, Tomas Pa- jdla, and Josef Sivic. NetVLAD: CNN architecture for weakly supervised place recognition. In CVPR, 2016. 3

work page 2016
[5]

RelocNet: Continuous metric learning relocalisation using neural nets

Vassileios Balntas, Shuda Li, and Victor Adrian Prisacariu. RelocNet: Continuous metric learning relocalisation using neural nets. In ECCV, 2018. 3

work page 2018
[6]

DSAC-Differentiable RANSAC for camera localiza- tion

Eric Brachmann, Alexander Krull, Sebastian Nowozin, Jamie Shotton, Frank Michel, Stefan Gumhold, and Carsten Rother. DSAC-Differentiable RANSAC for camera localiza- tion. In CVPR, 2017. 1, 2, 3, 4, 5, 9

work page 2017
[7]

Yang, Stefan Gumhold, and Carsten Rother

Eric Brachmann, Frank Michel, Alexander Krull, Michael Y . Yang, Stefan Gumhold, and Carsten Rother. Uncertainty- driven 6D pose estimation of objects and scenes from a sin- gle RGB image. In CVPR, 2016. 3

work page 2016
[8]

Learning less is more- 6D camera localization via 3D surface regression

Eric Brachmann and Carsten Rother. Learning less is more- 6D camera localization via 3D surface regression. In CVPR,

work page

Show all 54 references

[9]

Geometry-aware learning of maps for camera localization

Samarth Brahmbhatt, Jinwei Gu, Kihwan Kim, James Hays, and Jan Kautz. Geometry-aware learning of maps for camera localization. In CVPR, 2018. 3

2018
[10]

Random forests

Leo Breiman. Random forests. Machine Learning, 2001. 2

2001
[11]

Graph-based discriminative learning for location recognition

Song Cao and Noah Snavely. Graph-based discriminative learning for location recognition. In CVPR, 2013. 3

2013
[12]

On- the-ﬂy adaptation of regression forests for online camera re- localisation

Tommaso Cavallari, Stuart Golodetz, Nicholas A Lord, Julien Valentin, Luigi Di Stefano, and Philip HS Torr. On- the-ﬂy adaptation of regression forests for online camera re- localisation. In CVPR, 2017. 3

2017
[13]

Statistical Adjustment of Data

William Edwards Deming. Statistical Adjustment of Data
[14]

Fischler and Robert C

Martin A. Fischler and Robert C. Bolles. Random Sample Consensus: A paradigm for model ﬁtting with applications to image analysis and automated cartography. Commun. ACM,
[15]

Schapire

Yoav Freund and Robert E. Schapire. A short introduction to boosting. In IJCAI, 1999. 2 11

1999
[16]

Complete solution classiﬁcation for the perspective-three-point problem

Xiao-Shan Gao, Xiao-Rong Hou, Jianliang Tang, and Hang-Fei Cheng. Complete solution classiﬁcation for the perspective-three-point problem. TPAMI, 2003. 1, 6

2003
[17]

Multi-output learning for camera relocalization

Abner Guzman-Rivera, Pushmeet Kohli, Ben Glocker, Jamie Shotton, Toby Sharp, Andrew Fitzgibbon, and Shahram Izadi. Multi-output learning for camera relocalization. In CVPR, 2014. 3

2014
[18]

Deep residual learning for image recognition

Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In CVPR,
[19]

Hinton, Oriol Vinyals, and Jeffrey Dean

Geoffrey E. Hinton, Oriol Vinyals, and Jeffrey Dean. Distill- ing the knowledge in a neural network. In NIPS Workshops,
[20]

Jacobs, Michael I

Robert A. Jacobs, Michael I. Jordan, Steven J. Nowlan, and Geoffrey E. Hinton. Adaptive mixtures of local experts.Neu- ral Computation, 1991. 2, 3, 4

1991
[21]

A solution for the best rotation to re- late two sets of vectors

Wolfgang Kabsch. A solution for the best rotation to re- late two sets of vectors. Acta Crystallographica Section A: Crystal Physics, Diffraction, Theoretical and General Crys- tallography, 1976. 1

1976
[22]

Geometric loss functions for camera pose regression with deep learning

Alex Kendall and Roberto Cipolla. Geometric loss functions for camera pose regression with deep learning. In CVPR,
[23]

PoseNet: A convolutional network for real-time 6-DoF cam- era relocalization

Alex Kendall, Matthew Grimes, and Roberto Cipolla. PoseNet: A convolutional network for real-time 6-DoF cam- era relocalization. In ICCV, 2015. 3, 6

2015
[24]

Kingma and Jimmy Ba

Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In ICLR, 2015. 7

2015
[25]

Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton. ImageNet classiﬁcation with deep convolutional neural net- works. In NIPS, 2012. 2

2012
[26]

EPnP: An accurate O(n) solution to the PnP problem

Vincent Lepetit, Francesc Moreno-Noguer, and Pascal Fua. EPnP: An accurate O(n) solution to the PnP problem. IJCV,
[27]

Huttenlocher

Yunpeng Li, Noah Snavely, and Daniel P. Huttenlocher. Lo- cation recognition using prioritized feature matching. In ECCV, 2010. 8, 10

2010
[28]

Sinha, Michael F

Hyon Lim, Sudipta N. Sinha, Michael F. Cohen, and Matthew Uyttendaele. Real-time image-based 6-DoF local- ization in large-scale environments. In CVPR, 2012. 3

2012
[29]

Fully convolutional networks for semantic segmentation

Jonathan Long, Evan Shelhamer, and Trevor Darrell. Fully convolutional networks for semantic segmentation. In CVPR, 2015. 6

2015
[30]

Mixture of ex- perts: A literature survey

Saeed Masoudnia and Reza Ebrahimpour. Mixture of ex- perts: A literature survey. Artiﬁcial Intelligence Review ,
[31]

Daniela Massiceti, Alexander Krull, Eric Brachmann, Carsten Rother, and Philip H. S. Torr. Random forests ver- sus neural networks - What’s best for camera localization? In ICRA, 2017. 3

2017
[32]

Little, Julien Valentin, and Clarence W

Lili Meng, Jianhui Chen, Frederick Tung, James J. Little, Julien Valentin, and Clarence W. de Silva. Backtracking re- gression forests for accurate camera relocalization. In IROS,
[33]

Little, Julien Valentin, and Clarence W

Lili Meng, Frederick Tung, James J. Little, Julien Valentin, and Clarence W. de Silva. Exploiting points and lines in re- gression forests for RGB-D camera relocalization. In IROS,
[34]

Automatic differentiation in PyTorch

Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chanan, Edward Yang, Zachary DeVito, Zeming Lin, Al- ban Desmaison, Luca Antiga, and Adam Lerer. Automatic differentiation in PyTorch. In NIPS-W, 2017. 6

2017
[35]

Deep fundamental matrix estimation

Ren ´e Ranftl and Vladlen Koltun. Deep fundamental matrix estimation. In ECCV, 2018. 1

2018
[36]

Hyperpoints and ﬁne vocab- ularies for large-scale location recognition

Torsten Sattler, Michal Havlena, Filip Radenovic, Konrad Schindler, and Marc Pollefeys. Hyperpoints and ﬁne vocab- ularies for large-scale location recognition. In ICCV, 2015. 3

2015
[37]

Large-scale location recognition and the geomet- ric burstiness problem

Torsten Sattler, Michal Havlena, Konrad Schindler, and Marc Pollefeys. Large-scale location recognition and the geomet- ric burstiness problem. In CVPR, 2016. 3

2016
[38]

Efﬁcient & effective prioritized matching for large-scale image-based localization

Torsten Sattler, Bastian Leibe, and Leif Kobbelt. Efﬁcient & effective prioritized matching for large-scale image-based localization. TPAMI, 2016. 3, 9, 11

2016
[39]

Benchmarking 6DoF outdoor visual localiza- tion in changing conditions

Torsten Sattler, Will Maddern, Carl Toft, Akihiko Torii, Lars Hammarstrand, Erik Stenborg, Daniel Safari, Masatoshi Okutomi, Marc Pollefeys, Josef Sivic, Fredrik Kahl, and Tomas Pajdla. Benchmarking 6DoF outdoor visual localiza- tion in changing conditions. In CVPR, 2018. 8, 10

2018
[40]

Are large-scale 3D models really necessary for accurate visual localization? In CVPR, 2017

Torsten Sattler, Akihiko Torii, Josef Sivic, Marc Pollefeys, Hajime Taira, Masatoshi Okutomi, and Tomas Pajdla. Are large-scale 3D models really necessary for accurate visual localization? In CVPR, 2017. 3

2017
[41]

City-scale location recognition

Grant Schindler, Matthew Brown, and Richard Szeliski. City-scale location recognition. In CVPR, 2007. 3

2007
[42]

Estimating the dimension of a model

Gideon Schwarz. Estimating the dimension of a model. An- nals of Statistics, 1978. 3

1978
[43]

Le, Geoffrey E

Noam Shazeer, Azalia Mirhoseini, Krzysztof Maziarz, Andy Davis, Quoc V . Le, Geoffrey E. Hinton, and Jeff Dean. Outrageously large neural networks: The sparsely-gated mixture-of-experts layer. In ICLR, 2017. 2

2017
[44]

Scene co- ordinate regression forests for camera relocalization in RGB- D images

Jamie Shotton, Ben Glocker, Christopher Zach, Shahram Izadi, Antonio Criminisi, and Andrew Fitzgibbon. Scene co- ordinate regression forests for camera relocalization in RGB- D images. In CVPR, 2013. 3, 6, 7

2013
[45]

Very deep convo- lutional networks for large-scale image recognition

Karen Simonyan and Andrew Zisserman. Very deep convo- lutional networks for large-scale image recognition. CoRR,
[46]

InLoc: Indoor visual localization with dense matching and view synthesis

Hajime Taira, Masatoshi Okutomi, Torsten Sattler, Mircea Cimpoi, Marc Pollefeys, Josef Sivic, Tomas Pajdla, and Ak- ihiko Torii. InLoc: Indoor visual localization with dense matching and view synthesis. In CVPR, 2018. 3

2018
[47]

Seman- tic match consistency for long-term visual localization

Carl Toft, Erik Stenborg, Lars Hammarstrand, Lucas Brynte, Marc Pollefeys, Torsten Sattler, and Fredrik Kahl. Seman- tic match consistency for long-term visual localization. In ECCV, 2018. 3

2018
[48]

Learn- ing to navigate the energy landscape

Julien Valentin, Angela Dai, Matthias Nießner, Pushmeet Kohli, Philip Torr, Shahram Izadi, and Cem Keskin. Learn- ing to navigate the energy landscape. CoRR, 2016. 7

2016
[49]

Julien Valentin, Matthias Nießner, Jamie Shotton, Andrew Fitzgibbon, Shahram Izadi, and Philip H. S. Torr. Exploiting uncertainty in regression forests for accurate camera relocal- ization. In CVPR, 2015. 3

2015
[50]

Image- based localization with spatial LSTMs

Florian Walch, Caner Hazirbas, Laura Leal-Taix ´e, Torsten Sattler, Sebastian Hilsenbeck, and Daniel Cremers. Image- based localization with spatial LSTMs. In ICCV, 2017. 3 12

2017
[51]

HD-CNN: hierarchical deep con- volutional neural network for image classiﬁcation

Zhicheng Yan, Vignesh Jagadeesh, Dennis DeCoste, Wei Di, and Robinson Piramuthu. HD-CNN: hierarchical deep con- volutional neural network for image classiﬁcation. In ICCV,
[52]

Hierarchical mixture of classiﬁcation experts uncovers inter- actions between brain regions

Bangpeng Yao, Dirk Walther, Diane Beck, and Li Fei-fei. Hierarchical mixture of classiﬁcation experts uncovers inter- actions between brain regions. In NIPS, 2009. 2

2009
[53]

Learning to ﬁnd good correspondences

Kwang Moo Yi, Eduard Trulls, Yuki Ono, Vincent Lepetit, Mathieu Salzmann, and Pascal Fua. Learning to ﬁnd good correspondences. In CVPR, 2018. 1 13

2018
[2018]

3, 4, 6, 7, 9, 10, 11

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

Network of experts for large-scale image categoriza- tion

Karim Ahmed, Mohammad Haris Baig, and Lorenzo Torre- sani. Network of experts for large-scale image categoriza- tion. In ECCV, 2016. 2

work page 2016

[2] [2]

A new look at the statistical model identi- ﬁcation

Hirotugu Akaike. A new look at the statistical model identi- ﬁcation. TAC, 1974. 3

work page 1974

[3] [3]

Expert gate: Lifelong learning with a network of experts

Rahaf Aljundi, Punarjay Chakravarty, and Tinne Tuytelaars. Expert gate: Lifelong learning with a network of experts. In CVPR, 2017. 2, 3 Dubrovnik Aachen Figure 13. Ground Truth SfM Reconstructions. Both datasets contain outlier 3D points in the ground truth reconstruction. The outlier ratio is substantial for Dubrovnik, and still noticeable for Aachen

work page 2017

[4] [4]

NetVLAD: CNN architecture for weakly supervised place recognition

Relja Arandjelovi ´c, Petr Gronat, Akihiko Torii, Tomas Pa- jdla, and Josef Sivic. NetVLAD: CNN architecture for weakly supervised place recognition. In CVPR, 2016. 3

work page 2016

[5] [5]

RelocNet: Continuous metric learning relocalisation using neural nets

Vassileios Balntas, Shuda Li, and Victor Adrian Prisacariu. RelocNet: Continuous metric learning relocalisation using neural nets. In ECCV, 2018. 3

work page 2018

[6] [6]

DSAC-Differentiable RANSAC for camera localiza- tion

Eric Brachmann, Alexander Krull, Sebastian Nowozin, Jamie Shotton, Frank Michel, Stefan Gumhold, and Carsten Rother. DSAC-Differentiable RANSAC for camera localiza- tion. In CVPR, 2017. 1, 2, 3, 4, 5, 9

work page 2017

[7] [7]

Yang, Stefan Gumhold, and Carsten Rother

Eric Brachmann, Frank Michel, Alexander Krull, Michael Y . Yang, Stefan Gumhold, and Carsten Rother. Uncertainty- driven 6D pose estimation of objects and scenes from a sin- gle RGB image. In CVPR, 2016. 3

work page 2016

[8] [8]

Learning less is more- 6D camera localization via 3D surface regression

Eric Brachmann and Carsten Rother. Learning less is more- 6D camera localization via 3D surface regression. In CVPR,

work page