Pith. sign in

REVIEW 3 major objections 3 minor 83 references

DeepDISC-photoz: Deep Learning-Based Photometric Redshift Estimation for Rubin LSST

T0 review · 3 major / 3 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read On simulated Rubin images, a pixel-reading network beats catalog-based redshift estimators.

desk verdict A solid, openly-released image-based photo-z benchmark, but the headline outperformance claim is not yet fully controlled because the test set is DeepDISC-selected and BPZ runs with a known dust-law handicap. read the letter →

arxiv 2411.18769 v2 pith:EUGODYD7 submitted 2024-11-27 astro-ph.IM

Grant Merz , Xin Liu , Samuel Schmidt , Alex I. Malz , Tianqing Zhang , Doug Branton , Colin J. Burke , Melissa Delucchi
show 6 more authors
Yaswant Sai Ejjagiri Jeremy Kubica Yichen Liu Olivia Lynn Drew Oldag The LSST Dark Energy Science Collaboration
This is my paper · ORCID
classification astro-ph.IM
keywords photometricredshiftsdeeplearninginstancesegmentationmixturedensitynetworksRubinLSSTDC2simulationssourceblendingphoto-zPDFcalibration
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

Photometric redshifts, distances estimated from a few broad-band brightness measurements rather than spectra, are needed for billions of Rubin LSST galaxies. This paper claims that a single deep-learning network, an extension of the DeepDISC instance-segmentation framework, can detect, segment, classify, and estimate a full redshift probability distribution directly from multi-band images, skipping the usual step of building a photometric catalog. On simulated LSST (DC2) images, the network reports lower bias, scatter, and outlier fraction than the template-based code BPZ and the machine-learning code FlexZBoost, together with better-calibrated uncertainty distributions. The paper argues that the performance gain comes from pixel-level morphological and color-gradient information, and shows that the network is particularly robust to blended sources.

What carries the argument

The load-bearing object is the DeepDISC architecture with a Mixture Density Network (MDN) as the redshift head. The backbone is a multi-scale vision transformer (MViTv2) with a feature pyramid and cascade Region of Interest heads, trained end-to-end on six-band $ugrizy$ coadded images. For each detected object, the MDN outputs the weights, means, and log standard deviations of five Gaussian components, which parameterize the photo-z probability density function $p(z)$; the network is trained with the negative log-likelihood of the true redshifts. Milky Way dust reddening $E(B-V)$ enters as an extra neuron input to the MDN. During training, the detection and segmentation branches learn from deblended ground truth produced by the scarlet algorithm, while the redshift branch trains only on the LSST 'gold' sample ($i<25.3$ mag).

What would settle it

Run DeepDISC, BPZ, and FlexZBoost on a common test sample selected independently of DeepDISC (e.g., purely from the LSST science pipeline object catalog) and on images with realistic irregular galaxy morphologies and a corrected dust extinction law; if the deep-learning model's advantage in $\sigma_{\rm IQR}$ and outlier fraction $\eta$ shrinks or reverses under either change, the claimed outperformance is an artifact of the simulation or the selection scheme.

Watch

Extended reading notes

Core claim

The central claim is that adding a redshift-estimation Region of Interest head, implemented as a Mixture Density Network, to the DeepDISC object-detection framework yields a photo-z estimator that outperforms catalog-based estimators on simulated LSST data in both point estimates and probabilistic metrics. On the DC2 Year-5 test set, DeepDISC photo-z achieves bias $e_z=0.0007$, scatter $\sigma_{\rm IQR}=0.0412$, outlier fraction $\eta=0.1191$, and a CDE loss of $-4.249$, beating BPZ and FlexZBoost on every metric reported. The paper also establishes that photo-z scatter decreases roughly in proportion to the inverse of image signal-to-noise when comparing Year-1 and Year-5 coadds, and that pixel-level information is the carrier of the advantage: blurring the images degrades the model, and training FlexZBoost on DeepDISC's learned features recovers part of the gain over photometry.

Load-bearing premise

The benchmark assumes that the DC2 simulated images and the matched-catalog training and test scheme are a fair and representative stand-in for real LSST data; the paper itself notes that the simulated morphologies are simplified bulge+disk+knot profiles, that high-redshift SED diversity is limited, that a flaw in the simulated dust extinction law leaves BPZ with mismatched templates, and that the test set is selected by DeepDISC's own detections, which may privilege it.

Editorial extensions

If this is right

  • LSST photo-z production could become a single forward pass over coadded images that simultaneously detects, deblends, classifies, and assigns redshift PDFs, removing the separate forced-photometry catalog step.
  • Adding observing time improves photo-z scatter nearly proportionally to the gain in signal-to-noise, giving a quantitative forecast for how Year-1 to Year-10 coadds will improve redshift quality.
  • The strong robustness to blending implies that pixel-based estimators can partly bypass the hardest failure mode of catalog deblending, which should matter for weak-lensing shape and redshift analyses.
  • Secondary peaks in the DeepDISC PDFs carry genuine redshift information: evaluating the PDF at the secondary peak recovers a majority of the point-estimate outliers.
  • Since increasing model size and training-set size produced no clear gains, further progress is more likely to come from better pre-training or training-set augmentation than from scaling the current architecture.

Reading between the lines

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

  • If the claimed margin survives on real data, the practical implication is that the LSST photo-z pipeline could be simplified around one image-based estimator, reserving catalog-based codes as cross-checks; the paper's own selection caveat means this transfer must first be tested with an independent detection catalog.
  • The inverse-scatter-with-SNR scaling suggests a simple survey-design rule: photo-z quality in the i-band gold sample is set mostly by depth, so area-depth trade-offs can be evaluated before any new models are trained.
  • The feature-distillation experiment hints at a broader recipe: rather than discarding catalog methods, train them on features extracted by image models, which may offer a lower-cost hybrid with some of the image-based advantage; the paper only demonstrates this for FlexZBoost.
  • Testing on higher-fidelity simulations with irregular morphologies and a correct dust extinction law would sharpen the claim, since those are exactly the axes on which the paper admits the DC2 data depart from reality.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 3 minor

Summary. This paper presents DeepDISC-photoz, an extension of the DeepDISC instance-segmentation framework that adds a Mixture Density Network head producing photo-z PDFs directly from multi-band coadded images. Using the DESC DC2 simulations and the LSST gold sample, the authors train DeepDISC on 1048 cutouts and evaluate on 1925 test images, comparing against BPZ and FlexZBoost run on matched object-catalog photometry. They report that DeepDISC has lower bias, interquartile scatter, and outlier fraction, as well as better PIT calibration and lower CDE/KS/CvM/AD statistics, and that it is less sensitive to blending. The paper also examines depth dependence, model and training-set scaling, the information content of DeepDISC features versus photometry, stellar contamination, and the effect of image smoothing.

Significance. If the claims hold, this is a valuable step toward end-to-end image-based photo-z estimation for LSST, and the public code integrated in RAIL is a useful community resource. The paper's strengths include open release of code, candid disclosure of limitations, and a clean feature-versus-photometry control experiment. However, the headline comparison is weakened by the DeepDISC-selected test sample and by the acknowledged BPZ template mismatch, so the broad outperformance statement in the abstract is not yet fully supported by the presented evidence.

major comments (3)
  1. [Section 5.1 and Table 2] The global metrics in Table 2 are computed on a test set whose selection is determined by DeepDISC inference matched to the DC2 object catalog, as stated in Section 5.1. Because BPZ and FZB are evaluated only on objects that DeepDISC detected and cross-matched, a systematic detection preference could favor DeepDISC in every global metric. The Section 5.1 control retrains FZB on a DeepDISC-selected sample, but it does not re-run BPZ on a neutrally selected sample, and no global metrics are reported on a truth-selected sample that includes DeepDISC non-detections. Please add such an evaluation, or explicitly restrict the main outperformance claim to the DeepDISC-selected sample.
  2. [Section 3.2 and Section 3.2.1] The BPZ comparison is not fully controlled because the simulated dust extinction law is flawed and the BPZ template set cannot represent the resulting high-redshift colors; the paper states that this will degrade BPZ relative to Schmidt et al. (2020). Consequently, the global superiority of DeepDISC over BPZ in Table 2, and especially the advantage at 1.5 < z < 2.5 in Figure 5, may partly reflect template mismatch rather than estimator quality. Please either use a corrected template set, restrict the BPZ comparison to regimes where the templates are reliable, or clearly qualify the abstract's outperformance claim accordingly.
  3. [Section 4 and Section 5.3] DeepDISC never produces a PDF mode above z ~ 2.5, as the paper acknowledges in Section 5.3. This is a systematic failure in the high-redshift regime, yet Table 2 reports only global metrics. Because the abstract claims general outperformance in point-estimate metrics, the paper should quantify performance separately for z > 2.5 and explicitly state that the claimed advantage does not extend to that regime.
minor comments (3)
  1. [Section 4] The sentence 'At 1.5 ≤ z ≥ 2.5, DeepDISC maintains...' contains a typo; it should read '1.5 ≤ z ≤ 2.5'.
  2. [Table 2 caption] The caption 'DeepDISC outperforms BPZ and FZB in all cases' would be clearer as 'in all metrics', since the CDE loss is negative and the comparison is by magnitude rather than direction.
  3. [Table 2] Global metrics are quoted without uncertainties; bootstrapped errors are shown for binned metrics in Figure 5, but the global values in Table 2 should also include at least bootstrap or jackknife uncertainties.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: DeepDISC photo-z is benchmarked against independent catalog codes on simulated data; the acknowledged test-selection caveat is a validity concern, not a derivation reduced to its inputs.

full rationale

The central claim—that DeepDISC photo-z outperforms BPZ and FlexZBoost—is an empirical benchmark, not a derivation from its own outputs. Section 3.2 states 'we evaluate the performance of DeepDISC and compare to two catalog-based codes,' and all codes are scored with independently defined metrics (Equations 4-6 and 8-11) on a held-out test set. No DeepDISC parameter is fitted to the BPZ or FZB results, and no DeepDISC metric is defined in terms of the comparison codes' outputs. The main self-reference is to Merz et al. (2023) for the base architecture and transfer-learning choices; that prior work supplies the backbone and pretraining strategy, not the photo-z result, and the photo-z claim is tested against external codes rather than assumed. The paper also runs an independent ablation—'we compare an FZB model trained on photometry with an FZB model trained on the corresponding features produced by a trained DeepDISC model' (Section 5)—and a blurring ablation (Appendix) to support the pixel-information interpretation. The strongest caveat is sample selection: Section 5.1 states 'the test set selection is determined by DeepDISC inference matched with the DC2 object catalog, potentially privileging our method,' and Section 6 acknowledges 'We do not fully characterize the effects of object detection and unrecognized blends in this study.' These are disclosed correctness/external-validity limitations, not cases where a predicted quantity equals a fitted input by construction. Similarly, the paper's note that BPZ is handicapped by 'a flaw in the dust extinction law assumed in the simulations' (Section 3.2) is a fairness caveat, not circularity. No equation in the paper reduces the claimed outperformance to the training targets, and no load-bearing claim rests on an unverified self-citation. Therefore the circularity score is 0.

Assumptions & free parameters 6 free parameters · 7 assumptions · 0 invented entities

No free parameter is derived from the target result; the fitted quantities are network weights, architectural choices, and comparison-code hyperparameters. The paper introduces no new physical entities, mediators, forces, or dimensions. Its claims rest on the fidelity of DC2 simulations, the reliability of truth matching, and standard photo-z evaluation assumptions.

free parameters (6)
  • MDN Gaussian components = 5
    Adopted from D'Isanto & Polsterer (2018); the authors tested 3 and 7 components and saw no significant improvement (Section 3). This controls the expressive capacity of the photo-z PDF output.
  • MDN hidden layer sizes = 1024 and 64 units
    Redshift head architecture in Table 1; no external derivation is given. The capacity of this head affects how well complex PDFs can be represented.
  • MViTv2 backbone variant = Base
    Chosen following Merz et al. (2023); the scaling study later tests Large and Huge variants (Section 4.4). The backbone determines the features available to the redshift head.
  • Training schedule and learning rate = lr 0.001, drops by 10x at epochs 15, 25, 35; 50 epochs
    Hand-chosen training hyperparameters (Section 3). The reported performance depends on successful convergence under this schedule.
  • Detection IOU thresholds = [0.3, 0.7]
    Standard detectron2 thresholds for positive and negative detection matching (Section 3). They influence which objects enter the redshift head.
  • FlexZBoost hyperparameters = bumpmin=0.02, bumpmax=0.35, nbump=20, sharpmin=0.7, sharpmax=2.1, nsharp=15, max_depth=8
    RAIL defaults used for the baseline FlexZBoost comparison (Section 3.2.2). They affect the baseline performance against which DeepDISC is judged.
assumptions (7)
  • domain assumption DC2 simulated images are sufficiently realistic proxies for LSST imaging for relative algorithm comparison.
    The entire benchmark uses DC2 imSim images. The authors list limitations in Section 2, including simple bulge+disk+knot morphologies and limited high-redshift SED diversity, but still generalize conclusions toward LSST.
  • domain assumption Truth catalogs and scarlet-deblended masks provide correct training targets for detection, segmentation, and redshift.
    Supervised training assumes simulated truth is accurate; the authors note in Section 1 that biases in ground truth may propagate to inferred outputs.
  • domain assumption The catalog-to-truth matching rule (within 1 arcsec and r-band magnitude difference at most 1 mag, choosing the smallest magnitude difference) correctly associates objects with redshift labels.
    This matching is used to build training labels for BPZ and FlexZBoost and to produce the test cross-match (Section 3.2).
  • domain assumption The LSST gold sample, i-band magnitude < 25.3, is the relevant population for comparing photo-z estimators.
    The redshift head is trained only on gold-sample objects, following LSST science requirements (Sections 2 and 3).
  • domain assumption A five-component Gaussian mixture trained with negative log-likelihood can adequately represent photo-z PDFs.
    Adopted from D'Isanto & Polsterer (2018), with ad hoc tests of 3 and 7 components showing no improvement (Section 3).
  • domain assumption ImageNet-pretrained vision transformer weights transfer to six-band astronomical images.
    The transfer-learning strategy follows Merz et al. (2023). No fine-tuning of frozen layers is used because the channel count differs (Section 3).
  • standard math Standard photo-z evaluation metrics are valid for comparing estimators.
    The paper uses sigma_IQR, outlier fraction, PIT, CDE loss, KS, CvM, and AD statistics, which are standard in the photo-z literature (Schmidt & Malz 2020).

how reviews work

0 comments
Cite this review

Pith. "Pith review of DeepDISC-photoz: Deep Learning-Based Photometric Redshift Estimation for Rubin LSST." pith.science (2026). https://pith.science/paper/EUGODYD7

@misc{pith2026241118769,
  author       = {Pith},
  title        = {Pith review of: DeepDISC-photoz: Deep Learning-Based Photometric Redshift Estimation for Rubin LSST},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EUGODYD7}},
  note         = {Machine review of arXiv:2411.18769}
}
read the original abstract

Photometric redshifts will be a key data product for the Rubin Observatory Legacy Survey of Space and Time (LSST) as well as for future ground and space-based surveys. The need for photometric redshifts, or photo-zs, arises from sparse spectroscopic coverage of observed galaxies. LSST is expected to observe billions of objects, making it crucial to have a photo-z estimator that is accurate and efficient. To that end, we present DeepDISC photo-z, a photo-z estimator that is an extension of the DeepDISC framework. The base DeepDISC network simultaneously detects, segments, and classifies objects in multi-band coadded images. We introduce photo-z capabilities to DeepDISC by adding a redshift estimation Region of Interest head, which produces a photo-z probability distribution function for each detected object. On simulated LSST images, DeepDISC photo-z outperforms traditional catalog-based estimators, in both point estimate and probabilistic metrics. We validate DeepDISC by examining dependencies on systematics including galactic extinction, blending and PSF effects. We also examine the impact of the data quality and the size of the training set and model. We find that the biggest factor in DeepDISC photo-z quality is the signal-to-noise of the imaging data, and see a reduction in photo-z scatter approximately proportional to the image data signal-to-noise. Our code is fully public and integrated in the RAIL photo-z package for ease of use and comparison to other codes at https://github.com/LSSTDESC/rail_deepdisc

Figures

Figures reproduced from arXiv: 2411.18769 by the authors.

Figure 1
Figure 1. — An example DC2 image used for training the network. The RGB image corresponds to i, r and g bands and has been scaled with a Lupton asinh scaling (Lupton et al. 2004) for visualization purposes [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. — Histogram of the redshift distribution of objects in the training set. While the detection and segmentation branches of DeepDISC include all objects during training, the redshift estimation branch only trains using the magnitude-limited “gold" sample. the cosmoDC2 extragalactic catalog (Korytov et al. 2019) is created starting with an N-body simulation of 10,2403 par￾ticles in a volume of 4.225 Gpc3 . Simulations … view at source ↗
Figure 3
Figure 3. — Top: High-level diagram of the DeepDISC architecture. A multi-band image is input to the backbone network, which extracts features used for downstream tasks. The Region Proposal Network (RPN) is trained to identify which parts of the image contain an object. After these regions are identified, the corresponding features from the backbone are extracted and pooled together. The Region of Interest (ROI) heads then pe… view at source ↗
Figures from the paper (15 more)
Figure 4
Figure 4. Figure 4: — Redshift point estimates of DeepDISC (left), BPZ (middle) and FlexZBoost (right) compared to the true redshift. We use the mode of each photo-z PDF as our point estimate. The dashed line is along 𝑧true = 𝑧phot, and the solid lines define the 3𝜎IQR outlier boundary. C…
Figure 5
Figure 5. Figure 5: — Top: Histogram of galaxy redshifts in the test sample. Lower: Point estimate metrics as a function of true redshift. Error bars are derived from bootstrapping in each redshift bin. The LSST science requirements are shown as a gray band in the second panel and grey da…
Figure 6
Figure 6. Figure 6: — PIT histograms and QQ plots of the PDF ensembles produced by each code on the test set. PIT histograms are shown in grey and correspond to the right y-axis. The black horizontal line represents a uniform PIT histogram. The red line plots quantiles of the ensemble PIT…
Figure 7
Figure 7. Figure 7: — Recovery fraction, i.e., fraction of outliers that are no longer considered outliers if their PDF is evaluated at the secondary peak rather than the primary. We only include secondary peaks with a weight of at least 0.1 in the Figure, and plot the peak weight on the …
Figure 8
Figure 8. Figure 8: — Top: Histogram of Milky Way dust reddening E(B-V) for objects in the test sample. Lower: Bias, scatter, and outlier fraction metrics as a function of E(B-V). Error bars are derived from bootstrapping in each bin. The LSST science requirements are shown as a gray band…
Figure 9
Figure 9. Figure 9: — Top: Histogram of blendedness of objects in the test sample. Lower: Bias, scatter, and outlier fraction metrics as a function of blendedness. Error bars are derived from bootstrapping in each bin. The LSST science requirements are shown as a gray band in the second p…
Figure 10
Figure 10. Figure 10: — Top: Histogram of PSF FWHM in the i-band for objects in the test sample. Lower: Bias, scatter, and outlier fraction metrics as a function of PSF FWHM. Error bars are derived from bootstrapping in each bin. The LSST science requirements are shown as a gray band in th…
Figure 11
Figure 11. Figure 11: — DeepDISC photo-z estimates on objects in a Year 1 (left) and Year 5 (right) image. RGB colors correspond to the i,r, and g-band, respectively. A Lupton asinh scaling is used for visualization. A subset of randomly selected objects in the images is shown. The labels …
Figure 12
Figure 12. Figure 12: — Redshift point estimates for Year 1 (left) and Year 5 (right) data. The mode is used to produce a single point estimate for each pdf. The dashed line is along 𝑧true = 𝑧phot, and the solid lines define the 3𝜎IQR outlier boundary. Color corresponds to the number densi…
Figure 13
Figure 13. Figure 13: — Scaling laws for redshift point estimates. There appears to be no obvious relationship with model or training set size and point estimate metrics. The top panel shows the effect of increasing the training set size by 2x and 5x, and the bottom panel shows the effect …
Figure 14
Figure 14. Figure 14: — Scaling laws for redshift PDF estimates. The top panel shows the effect of increasing the training set size by 2x and 5x, and the bottom panel shows the effect of using larger model backbones, in our case the MViTv2 Base, Large, and Huge models. Similar to the point…
Figure 15
Figure 15. Figure 15: — After DeepDISC inference is ran, a small sample is set aside and divided into training and test samples for FZB. We compare the mode point estimate results on this small test sample. The dashed line is along 𝑧true = 𝑧phot, and the solid lines define the 3𝜎IQR outlie…
Figure 16
Figure 16. Figure 16: — PIT histograms and QQ plots of the PDF ensembles produced by DeepDISC (left), and FZB trained on photometry (middle) vs DeepDISC features (right). PIT histograms are shown in grey and correspond to the right y-axis. The black horizontal line represents a uniform PIT…
Figure 17
Figure 17. Figure 17: — Mode point estimates of the DeepDISC model trained with 5x the amount of training data. The network is able to better estimate photo-zs above 𝑧phot = 2.5. DISC outperforms FZB, but the metrics can in some sense be thought of as upper bounds on the performance. Mosko…
Figure 18
Figure 18. Figure 18: — The baseline DeepDISC model (left) compared to a model trained on blurred images (right). The model performs significantly worse on the blurred images, indicating that DeepDISC utilizes pixel-level information to extract more information about the sources than just …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

83 extracted references · 11 canonical work pages

  1. [1]

    Abbott T. M. C., et al., 2022, @doi [ ] 10.1103/PhysRevD.105.043512 , https://ui.adsabs.harvard.edu/abs/2022PhRvD.105d3512A 105, 043512

    doi:10.1103/physrevd.105.043512 2022

  2. [2]

    Aihara H., et al., 2018, @doi [Publications of the Astronomical Society of Japan] 10.1093/pasj/psx066 , 70, S4

    doi:10.1093/pasj/psx066 2018

  3. [3]

    W., Evrard A

    Allen S. W., Evrard A. E., Mantz A. B., 2011, @doi [ ] 10.1146/annurev-astro-081710-102514 , https://ui.adsabs.harvard.edu/abs/2011ARA&A..49..409A 49, 409

    doi:10.1146/annurev-astro-081710-102514 2011

  4. [4]

    Astropy Collaboration et al., 2013, @doi [ ] 10.1051/0004-6361/201322068 , http://adsabs.harvard.edu/abs/2013A

    doi:10.1051/0004-6361/201322068 2013

  5. [5]

    Ben \' tez N., 2000, @doi [ ] 10.1086/308947 , https://ui.adsabs.harvard.edu/abs/2000ApJ...536..571B 536, 571

    doi:10.1086/308947 2000

  6. [6]

    Aston University

    Bishop C., 1994, Working paper, Mixture density networks. Aston University

    1994

  7. [7]

    Bosch J., et al., 2018, @doi [Publications of the Astronomical Society of Japan] 10.1093/pasj/psx080 , 70, S5

    doi:10.1093/pasj/psx080 2018

  8. [8]

    Boucaud A., et al., 2020, @doi [ ] 10.1093/mnras/stz3056 , https://ui.adsabs.harvard.edu/abs/2020MNRAS.491.2481B 491, 2481

    doi:10.1093/mnras/stz3056 2020

Show all 83 references
  1. [9]

    J., Aleo P

    Burke C. J., Aleo P. D., Chen Y.-C., Liu X., Peterson J. R., Sembroski G. H., Lin J. Y.-Y., 2019, @doi [ ] 10.1093/mnras/stz2845 , https://ui.adsabs.harvard.edu/abs/2019MNRAS.490.3952B 490, 3952

    2019 doi

  2. [10]

    pp 6154--6162

    Cai Z., Vasconcelos N., 2018, in Proceedings of the IEEE conference on computer vision and pattern recognition. pp 6154--6162

    2018

  3. [11]

    Campagne J.-E., 2020, Adversarial training applied to Convolutional Neural Network for photometric redshift predictions ( @eprint arXiv 2002.10154 ), https://arxiv.org/pdf/2002.10154

    2020 arXiv

  4. [12]

    A., Clayton G

    Cardelli J. A., Clayton G. C., Mathis J. S., 1989, @doi [ ] 10.1086/167900 , https://ui.adsabs.harvard.edu/abs/1989ApJ...345..245C 345, 245

    1989 doi

  5. [13]

    arXiv:1603.02754

    Chen T., Guestrin C., 2016, @doi [arXiv e-prints] 10.48550/arXiv.1603.02754 , https://ui.adsabs.harvard.edu/abs/2016arXiv160302754C p. arXiv:1603.02754

    2016 doi arXiv

  6. [14]

    F., Kalmbach J

    Crenshaw J. F., Kalmbach J. B., Gagliano A., Yan Z., Connolly A. J., Malz A. I., Schmidt S. J., The LSST Dark Energy Science Collaboration 2024, @doi [ ] 10.3847/1538-3881/ad54bf , https://ui.adsabs.harvard.edu/abs/2024AJ....168...80C 168, 80

    2024 doi

  7. [15]

    DES 2018, @doi [ ] 10.1103/PhysRevD.98.043526 , https://ui.adsabs.harvard.edu/abs/2018PhRvD..98d3526A 98, 043526

    2018 doi

  8. [16]

    L., 2018, @doi [ ] 10.1051/0004-6361/201731326 , https://ui.adsabs.harvard.edu/abs/2018A&A...609A.111D 609, A111

    D'Isanto A., Polsterer K. L., 2018, @doi [ ] 10.1051/0004-6361/201731326 , https://ui.adsabs.harvard.edu/abs/2018A&A...609A.111D 609, A111

    2018 doi

  9. [17]

    B., Izbicki R., Freeman P

    Dalmasso N., Pospisil T., Lee A. B., Izbicki R., Freeman P. E., Malz A. I., 2020, @doi [Astronomy and Computing] 10.1016/j.ascom.2019.100362 , https://ui.adsabs.harvard.edu/abs/2020A&C....3000362D 30, 100362

    2020

  10. [18]

    Dark Energy Survey Collaboration et al., 2016, @doi [ ] 10.1093/mnras/stw641 , https://ui.adsabs.harvard.edu/abs/2016MNRAS.460.1270D 460, 1270

    2016 doi

  11. [19]

    A., Schneider M

    Dawson W. A., Schneider M. D., Tyson J. A., Jee M. J., 2016, @doi [ ] 10.3847/0004-637X/816/1/11 , http://adsabs.harvard.edu/abs/2016ApJ...816...11D 816, 11

    2016 doi

  12. [20]

    pp 248--255, @doi 10.1109/CVPR.2009.5206848

    Deng J., Dong W., Socher R., Li L.-J., Li K., Fei-Fei L., 2009, in 2009 IEEE Conference on Computer Vision and Pattern Recognition. pp 248--255, @doi 10.1109/CVPR.2009.5206848

    2009

  13. [21]

    H., Newman J

    Dey B., Andrews B. H., Newman J. A., Mao Y.-Y., Rau M. M., Zhou R., 2022, @doi [ ] 10.1093/mnras/stac2105 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.515.5285D 515, 5285

    2022 doi

  14. [22]

    T., 2003, @doi [ ] 10.1146/annurev.astro.41.011802.094840 , https://ui.adsabs.harvard.edu/abs/2003ARA&A..41..241D 41, 241

    Draine B. T., 2003, @doi [ ] 10.1146/annurev.astro.41.011802.094840 , https://ui.adsabs.harvard.edu/abs/2003ARA&A..41..241D 41, 241

    2003 arXiv

  15. [23]

    L., et al., 2015, @doi [ ] 10.1088/0004-637X/810/1/71 , https://ui.adsabs.harvard.edu/abs/2015ApJ...810...71F 810, 71

    Finkelstein S. L., et al., 2015, @doi [ ] 10.1088/0004-637X/810/1/71 , https://ui.adsabs.harvard.edu/abs/2015ApJ...810...71F 810, 71

    2015 doi

  16. [24]

    Gatti M., et al., 2021, @doi [ ] 10.1093/mnras/stab918 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.504.4312G 504, 4312

    2021 doi

  17. [25]

    E., Mu \ n oz R

    Gonz \'a lez R. E., Mu \ n oz R. P., Hern \'a ndez C. A., 2018, @doi [Astronomy and Computing] 10.1016/j.ascom.2018.09.004 , https://ui.adsabs.harvard.edu/abs/2018A&C....25..103G 25, 103

    2018 doi

  18. [26]

    Green G., 2018, @doi [The Journal of Open Source Software] 10.21105/joss.00695 , https://ui.adsabs.harvard.edu/abs/2018JOSS....3..695M 3, 695

    2018 doi

  19. [27]

    A., Stein G., Harrington P., Luki \'c Z., Mustafa M., 2021, @doi [ ] 10.3847/2041-8213/abf2c7 , https://ui.adsabs.harvard.edu/abs/2021ApJ...911L..33H 911, L33

    Hayat M. A., Stein G., Harrington P., Luki \'c Z., Mustafa M., 2021, @doi [ ] 10.3847/2041-8213/abf2c7 , https://ui.adsabs.harvard.edu/abs/2021ApJ...911L..33H 911, L33

    2021 doi

  20. [28]

    pp 2961--2969

    He K., Gkioxari G., Doll \'a r P., Girshick R., 2017, in Proceedings of the IEEE international conference on computer vision. pp 2961--2969

    2017

  21. [29]

    arXiv:1712.00409

    Hestness J., et al., 2017, @doi [arXiv e-prints] 10.48550/arXiv.1712.00409 , https://ui.adsabs.harvard.edu/abs/2017arXiv171200409H p. arXiv:1712.00409

    2017 doi arXiv

  22. [30]

    arXiv:1702.02600

    Huff E., Mandelbaum R., 2017, @doi [arXiv e-prints] 10.48550/arXiv.1702.02600 , https://ui.adsabs.harvard.edu/abs/2017arXiv170202600H p. arXiv:1702.02600

    2017 doi arXiv

  23. [31]

    D., 2007, @doi [Computing in Science & Engineering] 10.1109/MCSE.2007.55 , 9, 90

    Hunter J. D., 2007, @doi [Computing in Science & Engineering] 10.1109/MCSE.2007.55 , 9, 90

    2007 doi

  24. [32]

    Ivezi \'c Z ., et al., 2019, @doi [ ] 10.3847/1538-4357/ab042c , https://ui.adsabs.harvard.edu/abs/2019ApJ...873..111I 873, 111

    2019 doi

  25. [33]

    B., 2017, @doi [Electronic Journal of Statistics] 10.1214/17-EJS1302 , 11, 2800

    Izbicki R., Lee A. B., 2017, @doi [Electronic Journal of Statistics] 10.1214/17-EJS1302 , 11, 2800

    2017 doi

  26. [34]

    Jenness T., et al., 2022, in Software and Cyberinfrastructure for Astronomy VII. p. 1218911 ( @eprint arXiv 2206.14941 ), @doi 10.1117/12.2629569

    2022 arXiv

  27. [35]

    Korytov D., et al., 2019, @doi [ ] 10.3847/1538-4365/ab510c , https://ui.adsabs.harvard.edu/abs/2019ApJS..245...26K 245, 26

    2019 doi

  28. [36]

    LSST-DESC PZ WG 2023, LSSTDESC/RAIL: v0.98.5, @doi 10.5281/zenodo.7927358 , https://doi.org/10.5281/zenodo.7927358

    2023 doi

  29. [37]

    LSST Dark Energy Science Collaboration (LSST DESC) et al., 2021, @doi [ ] 10.3847/1538-4365/abd62c , https://ui.adsabs.harvard.edu/abs/2021ApJS..253...31L 253, 31

    2021 doi

  30. [38]

    arXiv:0912.0201

    LSST Science Collaboration et al., 2009, @doi [arXiv e-prints] 10.48550/arXiv.0912.0201 , https://ui.adsabs.harvard.edu/abs/2009arXiv0912.0201L p. arXiv:0912.0201

    2009 doi arXiv

  31. [39]

    arXiv:2310.03024

    Lanusse F., et al., 2023, @doi [arXiv e-prints] 10.48550/arXiv.2310.03024 , https://ui.adsabs.harvard.edu/abs/2023arXiv231003024L p. arXiv:2310.03024

    2023 doi arXiv

  32. [40]

    pp 4804--4814

    Li Y., Wu C.-Y., Fan H., Mangalam K., Xiong B., Malik J., Feichtenhofer C., 2022, in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp 4804--4814

    2022

  33. [41]

    arXiv:1612.03144

    Lin T.-Y., Doll \'a r P., Girshick R., He K., Hariharan B., Belongie S., 2016, @doi [arXiv e-prints] 10.48550/arXiv.1612.03144 , https://ui.adsabs.harvard.edu/abs/2016arXiv161203144L p. arXiv:1612.03144

    2016 doi arXiv

  34. [42]

    pp 10012--10022

    Liu Z., Lin Y., Cao Y., Hu H., Wei Y., Zhang Z., Lin S., Guo B., 2021, in Proceedings of the IEEE/CVF international conference on computer vision. pp 10012--10022

    2021

  35. [43]

    R., Fekete G., Hogg D

    Lupton R., Blanton M. R., Fekete G., Hogg D. W., O'Mullane W., Szalay A., Wherry N., 2004, @doi [ ] 10.1086/382245 , https://ui.adsabs.harvard.edu/abs/2004PASP..116..133L 116, 133

    2004 doi

  36. [44]

    Mandelbaum R., 2018, @doi [Annual Review of Astronomy and Astrophysics] https://doi.org/10.1146/annurev-astro-081817-051928 , 56, 393

    2018 doi

  37. [45]

    Massarotti M., Iovino A., Buzzoni A., 2001, @doi [ ] 10.1051/0004-6361:20000553 , https://ui.adsabs.harvard.edu/abs/2001A&A...368...74M 368, 74

    2001 doi

  38. [46]

    Melchior P., Moolekamp F., Jerdee M., Armstrong R., Sun A.-L., Bosch J., Lupton R., 2018, @doi [Astronomy and Computing] 10.1016/j.ascom.2018.07.001 , 24, 129

    2018 doi

  39. [47]

    Melchior P., Joseph R., Sanchez J., MacCrann N., Gruen D., 2021, @doi [Nat Rev Phys] 10.1038/s42254-021-00353-y , 3, 712

    2021 doi

  40. [48]

    J., Aleo P

    Merz G., Liu Y., Burke C. J., Aleo P. D., Liu X., Carrasco Kind M., Kindratenko V., Liu Y., 2023, @doi [ ] 10.1093/mnras/stad2785 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.526.1122M 526, 1122

    2023 doi

  41. [49]

    Mitra A., Kessler R., More S., Hlozek R., LSST Dark Energy Science Collaboration 2023, @doi [ ] 10.3847/1538-4357/acb057 , https://ui.adsabs.harvard.edu/abs/2023ApJ...944..212M 944, 212

    2023 doi

  42. [50]

    Miyatake H., et al., 2023, @doi [ ] 10.1103/PhysRevD.108.123517 , https://ui.adsabs.harvard.edu/abs/2023PhRvD.108l3517M 108, 123517

    2023 doi

  43. [51]

    F., Andrews B

    Moskowitz I., Gawiser E., Crenshaw J. F., Andrews B. H., Malz A. I., Schmidt S., LSST Dark Energy Science Collaboration 2024, @doi [ ] 10.3847/2041-8213/ad4039 , https://ui.adsabs.harvard.edu/abs/2024ApJ...967L...6M 967, L6

    2024 doi

  44. [52]

    D., White M., Ball N

    Myers A. D., White M., Ball N. M., 2009, @doi [ ] 10.1111/j.1365-2966.2009.15432.x , https://ui.adsabs.harvard.edu/abs/2009MNRAS.399.2279M 399, 2279

    2009

  45. [53]

    Myles J., et al., 2021, @doi [ ] 10.1093/mnras/stab1515 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.505.4249M 505, 4249

    2021 doi

  46. [54]

    A., Gruen D., 2022, @doi [ ] 10.1146/annurev-astro-032122-014611 , https://ui.adsabs.harvard.edu/abs/2022ARA&A..60..363N 60, 363

    Newman J. A., Gruen D., 2022, @doi [ ] 10.1146/annurev-astro-032122-014611 , https://ui.adsabs.harvard.edu/abs/2022ARA&A..60..363N 60, 363

    2022 doi

  47. [55]

    Oldag D., et al., 2024, @doi [Research Notes of the AAS] 10.3847/2515-5172/ad4da1 , 8, 141

    2024 doi

  48. [56]

    Pasquet J., Bertin E., Treyer M., Arnouts S., Fouchez D., 2019, @doi [ ] 10.1051/0004-6361/201833617 , https://ui.adsabs.harvard.edu/abs/2019A&A...621A..26P 621, A26

    2019 doi

  49. [57]

    Paszke A., et al., 2017, in NIPS-W

    2017

  50. [58]

    J., et al., 2024, @doi [RAS Techniques and Instruments] 10.1093/rasti/rzae013 , https://ui.adsabs.harvard.edu/abs/2024RASTI...3..174P 3, 174

    Popp J. J., et al., 2024, @doi [RAS Techniques and Instruments] 10.1093/rasti/rzae013 , https://ui.adsabs.harvard.edu/abs/2024RASTI...3..174P 3, 174

    2024 doi

  51. [59]

    M., et al., 2018, @doi [ ] 10.3847/1538-3881/aabc4f , https://ui.adsabs.harvard.edu/#abs/2018AJ....156..123T 156, 123

    Price-Whelan A. M., et al., 2018, @doi [ ] 10.3847/1538-3881/aabc4f , https://ui.adsabs.harvard.edu/#abs/2018AJ....156..123T 156, 123

    2018 doi

  52. [60]

    D., Lee D

    Ren S., He K., Girshick R., Sun J., 2015, in Cortes C., Lawrence N. D., Lee D. D., Sugiyama M., Garnett R., eds, , Advances in Neural Information Processing Systems 28. Curran Associates, Inc., pp 91--99

    2015

  53. [61]

    arXiv:2411.07305

    Roster W., et al., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2411.07305 , https://ui.adsabs.harvard.edu/abs/2024arXiv241107305R p. arXiv:2411.07305

    2024 doi arXiv

  54. [62]

    Rowe B. T. P., et al., 2015, @doi [Astronomy and Computing] 10.1016/j.ascom.2015.02.002 , https://ui.adsabs.harvard.edu/abs/2015A&C....10..121R 10, 121

    2015 doi

  55. [63]

    Salvato M., Ilbert O., Hoyle B., 2019, @doi [Nature Astronomy] 10.1038/s41550-018-0478-0 , https://ui.adsabs.harvard.edu/abs/2019NatAs...3..212S 3, 212

    2019 doi

  56. [64]

    F., Finkbeiner D

    Schlafly E. F., Finkbeiner D. P., 2011, @doi [ ] 10.1088/0004-637X/737/2/103 , https://ui.adsabs.harvard.edu/abs/2011ApJ...737..103S 737, 103

    2011 doi

  57. [65]

    J., et al., 2020, @doi [ ] 10.1093/mnras/staa2799 , https://ui.adsabs.harvard.edu/abs/2020MNRAS.499.1587S 499, 1587

    Schmidt S. J., et al., 2020, @doi [ ] 10.1093/mnras/staa2799 , https://ui.adsabs.harvard.edu/abs/2020MNRAS.499.1587S 499, 1587

    2020 doi

  58. [66]

    H., Ca \ n ameras R., Taubenberger S., Meinhardt T., Leal-Taix \'e L., Hsieh B

    Schuldt S., Suyu S. H., Ca \ n ameras R., Taubenberger S., Meinhardt T., Leal-Taix \'e L., Hsieh B. C., 2021, @doi [ ] 10.1051/0004-6361/202039945 , https://ui.adsabs.harvard.edu/abs/2021A&A...651A..55S 651, A55

    2021 doi

  59. [67]

    Seo H.-J., et al., 2012, @doi [ ] 10.1088/0004-637X/761/1/13 , https://ui.adsabs.harvard.edu/abs/2012ApJ...761...13S 761, 13

    2012 doi

  60. [68]

    S., Huff E

    Sheldon E. S., Huff E. M., 2017, @doi [ ] 10.3847/1538-4357/aa704b , https://ui.adsabs.harvard.edu/abs/2017ApJ...841...24S 841, 24

    2017 doi

  61. [69]

    S., Becker M

    Sheldon E. S., Becker M. R., MacCrann N., Jarvis M., 2020, @doi [ ] 10.3847/1538-4357/abb595 , https://ui.adsabs.harvard.edu/abs/2020ApJ...902..138S 902, 138

    2020 doi

  62. [70]

    J., Roberts R

    Smith M. J., Roberts R. J., Angeloudi E., Huertas-Company M., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2405.14930 , https://ui.adsabs.harvard.edu/abs/2024arXiv240514930S p. arXiv:2405.14930

    2024 doi arXiv

  63. [71]

    Soo J. Y. H., et al., 2018, @doi [ ] 10.1093/mnras/stx3201 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.475.3613S 475, 3613

    2018 doi

  64. [72]

    K., 2024, @doi [Monthly Notices of the Royal Astronomical Society] 10.1093/mnras/stae411 , 530, 2012

    Tanigawa S., Glazebrook K., Jacobs C., Labbe I., Qin A. K., 2024, @doi [Monthly Notices of the Royal Astronomical Society] 10.1093/mnras/stae411 , 530, 2012

    2024 doi

  65. [73]

    Treu T., 2010, @doi [ ] 10.1146/annurev-astro-081309-130924 , https://ui.adsabs.harvard.edu/abs/2010ARA&A..48...87T 48, 87

    2010 doi

  66. [74]

    Treyer M., Ait Ouahmed R., Pasquet J., Arnouts S., Bertin E., Fouchez D., 2024, @doi [ ] 10.1093/mnras/stad3171 , https://ui.adsabs.harvard.edu/abs/2024MNRAS.527..651T 527, 651

    2024 doi

  67. [75]

    A., et al., 2018, @doi [ ] 10.1103/PhysRevD.98.043528 , https://ui.adsabs.harvard.edu/abs/2018PhRvD..98d3528T 98, 043528

    Troxel M. A., et al., 2018, @doi [ ] 10.1103/PhysRevD.98.043528 , https://ui.adsabs.harvard.edu/abs/2018PhRvD..98d3528T 98, 043528

    2018 doi

  68. [76]

    A., et al., 2023, @doi [ ] 10.1093/mnras/stad664 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.522.2801T 522, 2801

    Troxel M. A., et al., 2023, @doi [ ] 10.1093/mnras/stad664 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.522.2801T 522, 2801

    2023 doi

  69. [77]

    arXiv:2404.02973

    Walmsley M., et al., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2404.02973 , https://ui.adsabs.harvard.edu/abs/2024arXiv240402973W p. arXiv:2404.02973

    2024 doi arXiv

  70. [78]

    L., Han J

    Wen Z. L., Han J. L., 2021, @doi [ ] 10.1093/mnras/staa3308 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.500.1003W 500, 1003

    2021 doi

  71. [79]

    Wilson D., Nayyeri H., Cooray A., H \"a u ler B., 2020, @doi [ ] 10.3847/1538-4357/ab5a79 , https://ui.adsabs.harvard.edu/abs/2020ApJ...888...83W 888, 83

    2020 doi

  72. [80]

    Wu Y., Kirillov A., Massa F., Lo W.-Y., Girshick R., 2019, Detectron2, https://github.com/facebookresearch/detectron2

    2019

  73. [81]

    arXiv:2106.04560

    Zhai X., Kolesnikov A., Houlsby N., Beyer L., 2021, @doi [arXiv e-prints] 10.48550/arXiv.2106.04560 , https://ui.adsabs.harvard.edu/abs/2021arXiv210604560Z p. arXiv:2106.04560

    2021 doi arXiv

  74. [82]

    pp 12104--12113

    Zhai X., Kolesnikov A., Houlsby N., Beyer L., 2022, in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp 12104--12113

    2022

  75. [83]

    Zhang Y.-C., Yang X.-H., 2019, @doi [Research in Astronomy and Astrophysics] 10.1088/1674-4527/19/1/6 , https://ui.adsabs.harvard.edu/abs/2019RAA....19....6Z 19, 006

    2019 doi

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.