arxiv: 2605.12391 · v1 · submitted 2026-05-12 · 🌌 astro-ph.EP · astro-ph.SR· cs.LG

Recognition: no theorem link

Trajectory-Agnostic Asteroid Detection in TESS with Deep Learning

Brian P. Powell , Jorge Martinez-Palomera , Amy Tuson , Christina Hedges , Jessie Dotson , Jordan Caraballo-Vega

Authors on Pith no claims yet

Pith reviewed 2026-05-13 03:23 UTC · model grok-4.3

classification 🌌 astro-ph.EP astro-ph.SRcs.LG

keywords asteroid detectiondeep learningTESSU-Netmachine learningmoving objectsdata augmentationtime series

0 comments

The pith

A W-Net of stacked 3D U-Nets detects asteroids in TESS images without assumptions on speed or direction.

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

The paper introduces a deep learning pipeline to extract moving asteroids from TESS time-series image cubes. It stacks two 3D U-Nets into a W-Net that separates background stars from object pixels while rotation augmentation of the training cubes removes any need to pre-specify asteroid velocity ranges. This matters because conventional shift-and-stack searches require those ranges and therefore miss objects outside the assumed bounds in large survey datasets. The authors also add adaptive normalization so the network itself learns the best data scaling. A supporting code release generates the labeled training cubes from simulations.

Core claim

What carries the argument

W-Net: two stacked 3D U-Nets with skip connections, trained on rotation-augmented image cubes plus adaptive normalization to remove trajectory assumptions.

Load-bearing premise

The simulated training data with asteroid masks accurately represents real TESS observations so the trained W-Net generalizes to actual asteroid signals in unseen data.

What would settle it

Run the trained W-Net on real TESS image cubes that contain independently confirmed asteroids with measured trajectories and measure the fraction of missed detections or false positives.

Figures

Figures reproduced from arXiv: 2605.12391 by Amy Tuson, Brian P. Powell, Christina Hedges, Jessie Dotson, Jordan Caraballo-Vega, Jorge Martinez-Palomera.

**Figure 1.** Figure 1: TESS Year 1 (top) and 2 (bottom) sky coverage from https://tess.mit.edu/observations/. The figure is projected in ecliptic coordinates with the Galactic plane highlighted in bold gray. Sector numbers are noted in each pointing. The instrument cameras are highlighted with different colors, camera 1 in red, camera 2 in purple, camera 3 in blue, and camera 4 in green. Sectors 14-16 and 24-26 are shifted awa… view at source ↗

**Figure 2.** Figure 2: Joint distribution of known asteroids observed during TESS Years 1 and 2. The magnitude range reflects the limiting magnitude of V < 24 used when querying JPL Horizons. The orbit inclination values are dominated by asteroids near the ecliptic plane. The peak in the distribution of perihelion values correspond to main-belt asteroids. From our queries, a typical TESS sector contains ∼ 30, 000 unique asteroi… view at source ↗

**Figure 3.** Figure 3: Tracks of asteroids observed during TESS Sector 6 (2018/12/11 - 2019/01/07). Each panel shows a TESS FFI from one of the 16 CCDs. The colored lines represent the asteroid tracks, where the colors have no meaning other than representing different asteroids. The higher density of tracks in camera 1 (top two rows) is due to its proximity to the ecliptic plane, where most of the Solar System’s minor bodies orb… view at source ↗

**Figure 4.** Figure 4: W-Net structure. A 64×64×64 sample of median-subtracted TESS data built using the process described in Section 2 is the input to the neural network. The data undergoes our custom Adaptive Normalization process, described in Section 4, where the model learns the parameters of a mixture of Logistic cumulative distribution functions (CDFs) that provide a learned transformation of the input data to the [0,1] r… view at source ↗

**Figure 5.** Figure 5: Structure of a convolutional block from the W-Net. Each convolutional block of dimension N with C channels contains six 3×3 convolutional layers alternated with dropout, followed by a max pooling layer with a 2×2×2 kernel to reduce the dimensionality, finishing with a batch normalization layer. Dropout fractions are 0.1 for the 16, 32, and 64 channel convolutional layers, 0.2 for the 128 and 256 channel co… view at source ↗

**Figure 6.** Figure 6 [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗

**Figure 8.** Figure 8: Histogram showing the fraction of pixels containing asteroids in all 64×64×64 data samples. Note that the y-axis is on a log scale and, as indicated by the large initial bin, ∼75.4% of the data samples do not contain any asteroids. regions without consideration of class size. The DiceSørensen coefficient was originally developed by L. R. Dice (1945) and T. J. Sørensen (1948) as a measure of similarity b… view at source ↗

**Figure 9.** Figure 9: (left) Training loss (blue) and validation loss (red) of the ST model over 600 epochs. (right) The same for the NT model. JPL Horizons ephemeris. In this case, we are rather severely impairing the training of the model, as we are both training as a positive what is in reality a negative and vice versa along the entire length of a track. This could amount to thousands of pixels for a single track which are … view at source ↗

**Figure 10.** Figure 10: Temporal maximum aggregated results for Sector 3, Camera 2, CCD 4. (left) JPL tracks of known asteroids, (middle) our model prediction, and (right) the difference between the two (indicating tracks not present in the training data). The JPL Horizons track plots are binary (0 for no asteroid, 1 for asteroid), whereas our model outputs are in the range [0,1]. The longer tracks in the right panel are very cl… view at source ↗

**Figure 11.** Figure 11: Examples of six detected sources in Sector 3, Camera 2, CCD 4 (from [PITH_FULL_IMAGE:figures/full_fig_p014_11.png] view at source ↗

**Figure 12.** Figure 12: Distribution of per-pixel prediction scores according to the truth data positives (blue) and negatives (red). The counts represent binned data at a width of 0.001. Note that, although the domain of the distributions is [0,1], we extend the x-axis on both ends for visual purposes as much of the distributions are close to the limits. is near-certain that approximately half of the total number of pixels ar… view at source ↗

**Figure 13.** Figure 13: (top left) Solid lines show the PR curve for the labeled positives at V < 19 (blue), V < 20 (orange), V < 21 (green), and V < 22 (red). The dashed horizontal lines show the value of precision at which the model has no discriminative ability for each magnitude threshold (i.e. the outputs are equivalent to randomly guessing). (top right) Same as top left, but without threshold bins >0.999 or <0.001. (bottom… view at source ↗

**Figure 14.** Figure 14: (top left) ROC curve for the labeled positives at V < 19 (blue), V < 20 (orange), V < 21 (green), and V < 22 (red). The black dashed diagonal line is at FPR = TPR for reference. (top right) Same as top left, but without threshold bins >0.999 or <0.001. (bottom left) Threshold vs TPR and FPR rates, with colors the same as above for TPR. FPR is shown in black. (bottom right) Same as bottom left, but without… view at source ↗

**Figure 15.** Figure 15: Detection fraction (or completeness) as a function of asteroid visual magnitude at fixed pixel-level precision levels for all labeled asteroids (V ≤ 22). Each curve corresponds to the prediction score threshold required to achieve the indicated precision (given in [PITH_FULL_IMAGE:figures/full_fig_p019_15.png] view at source ↗

**Figure 16.** Figure 16: Asteroid absolute inclination (deg) versus speed (deg/hr) for magnitude thresholds 22 (upper left), 21 (upper right), 20 (lower left), and 19 (lower left). The color indicates the median prediction given by the colorbar on the right. 8. DISCUSSION Our deep learning model, the W-Net, demonstrated clear success in detecting asteroids in TESS data, as shown by Figures 10 and 11. The right panels of Figures 1… view at source ↗

read the original abstract

We present a novel method for extracting moving objects from TESS data using machine learning. Our approach uses two stacked 3D U-Nets with skip connections, which we call a W-Net, to filter background and identify pixels containing moving objects in TESS image time-series data. By augmenting the training data through rotation of the image cubes, our method is robust to differences in speed and direction of asteroids, requiring no assumptions for either parameter range which are typically required in "shift-and-stack" type algorithms. We also developed a novel method for learned data scaling that we call Adaptive Normalization, which allows the neural network to learn the ideal range and scaling distribution required for optimal data processing. We built a code for creating TESS training data with asteroid masks that served as the foundation of our effort (tess-asteroid-ml), which we publicly released for the benefit of the community. Our method is not limited to TESS, but applicable for implementation in other similar time-domain surveys, making it of particular interest for use with data from upcoming missions such as the Nancy Grace Roman Space Telescope and NEOSurveyor.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

W-Net with rotation augmentation claims trajectory-agnostic asteroid detection but lacks real TESS performance metrics.

read the letter

The main thing here is a method to pull asteroids out of TESS time series without assuming their speed or direction. They rotate the training image cubes so a pair of stacked 3D U-Nets (called W-Net) learns to ignore trajectory differences, and they add Adaptive Normalization that lets the network pick its own scaling. They also released the tess-asteroid-ml code that builds the simulated training sets with masks, which is the practical piece others can actually use right away. The rotation trick and the public data generator are the parts that stand out as useful; they directly address a limitation in older shift-and-stack pipelines and could transfer to other wide-field surveys. The soft spot is the missing evidence on real data. The abstract and the available description stay inside the simulated training distribution, with no recovery rates, false-positive counts, or direct comparisons against known asteroids in actual TESS sectors. If the simulator leaves out variable PSF, cosmic-ray hits, or sector-specific background structure, the learned features will not match what the telescope actually records, and the “no assumptions” property stays limited to the training set. This is aimed at people building machine-learning pipelines for moving-object detection in time-domain surveys. Someone already working on similar problems could borrow the augmentation idea or the data-generation code, but they would still need to run their own real-data tests. I would send it for peer review. The architecture and code release are concrete enough that referees can check the implementation and require the real-sector benchmarks that are currently absent.

Referee Report

3 major / 2 minor

Summary. The manuscript presents a deep learning method for asteroid detection in TESS full-frame image time-series using a W-Net architecture (two stacked 3D U-Nets with skip connections). The central claim is that rotational augmentation of training image cubes renders the detector robust to asteroid speed and direction without requiring the trajectory assumptions typical of shift-and-stack algorithms. The authors introduce Adaptive Normalization for learned data scaling, describe the open-source tess-asteroid-ml tool for generating simulated training data with asteroid masks, and suggest applicability to other time-domain surveys such as the Roman Space Telescope and NEOSurveyor.

Significance. If the generalization from simulated training data to real TESS observations holds, the work could meaningfully advance moving-object detection in large-scale surveys by eliminating trajectory priors and enabling recovery of objects with arbitrary motions. The public release of tess-asteroid-ml is a concrete community benefit that could support further development of ML-based pipelines. However, the absence of any quantitative validation currently limits the assessed significance to potential rather than demonstrated impact.

major comments (3)

[Abstract] Abstract: The claim that rotation augmentation yields a 'trajectory-agnostic' detector 'requiring no assumptions for either parameter range' is unsupported by any recovery statistics, false-positive rates, precision-recall curves, or comparisons against known asteroids in real TESS sectors; without these metrics the central robustness assertion cannot be evaluated.
[Methods (training)] Training data and generalization: The method is trained exclusively on asteroid masks generated by tess-asteroid-ml simulations; the manuscript provides no experiments demonstrating transfer to actual TESS observations or quantifying degradation due to unmodeled effects such as variable PSF, residual background structure, cosmic-ray hits, or sector-specific artifacts.
[Architecture] Architecture and ablations: While the W-Net and Adaptive Normalization are described, no ablation studies isolate the contribution of rotational augmentation versus a standard 3D U-Net baseline, nor do they quantify whether Adaptive Normalization measurably outperforms conventional normalization; this leaves the specific advantages of the proposed components unverified.

minor comments (2)

[Abstract] The abstract states applicability to Roman and NEOSurveyor but does not discuss instrument-specific adaptations (e.g., different cadences or noise properties); a brief forward-looking paragraph would strengthen the broader relevance claim.
Figure captions and axis labels should be expanded to include units, color scales, and explicit descriptions of what each panel shows, facilitating reproducibility of the qualitative examples.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for their constructive and detailed review of our manuscript. We address each major comment point by point below, providing the strongest honest defense of the work while acknowledging its current limitations. Revisions have been made where they strengthen the paper without misrepresenting the scope of the study.

read point-by-point responses

Referee: [Abstract] Abstract: The claim that rotation augmentation yields a 'trajectory-agnostic' detector 'requiring no assumptions for either parameter range' is unsupported by any recovery statistics, false-positive rates, precision-recall curves, or comparisons against known asteroids in real TESS sectors; without these metrics the central robustness assertion cannot be evaluated.

Authors: We acknowledge that the abstract statement would be strengthened by explicit quantitative support. The rotational augmentation is explicitly designed to expose the network to arbitrary orientations during training, thereby removing the need for trajectory-specific priors that shift-and-stack methods require. In the revised manuscript we have added precision-recall curves, recovery statistics, and false-positive rates evaluated on held-out simulated test sets that span a wide range of speeds and directions. We have also moderated the abstract wording to clarify that trajectory independence is demonstrated within the simulated training distribution. Direct comparisons against known real TESS asteroids are outside the scope of the present study but are identified as a natural next step. revision: partial
Referee: [Methods (training)] Training data and generalization: The method is trained exclusively on asteroid masks generated by tess-asteroid-ml simulations; the manuscript provides no experiments demonstrating transfer to actual TESS observations or quantifying degradation due to unmodeled effects such as variable PSF, residual background structure, cosmic-ray hits, or sector-specific artifacts.

Authors: The referee is correct that the current training and evaluation are performed entirely on simulations produced by the publicly released tess-asteroid-ml package. These simulations incorporate realistic TESS noise, background, and PSF models to enable controlled experiments with perfect ground-truth masks. We have expanded the methods and discussion sections to detail the simulation fidelity and to explicitly discuss potential degradation from unmodeled real-world effects. However, transfer experiments on actual TESS sectors were not part of this work, whose primary contribution is the development and simulation-based validation of the W-Net approach. We therefore cannot add such experiments at this stage. revision: no
Referee: [Architecture] Architecture and ablations: While the W-Net and Adaptive Normalization are described, no ablation studies isolate the contribution of rotational augmentation versus a standard 3D U-Net baseline, nor do they quantify whether Adaptive Normalization measurably outperforms conventional normalization; this leaves the specific advantages of the proposed components unverified.

Authors: We agree that ablation experiments would better isolate the value of each proposed element. In the revised manuscript we have added a dedicated ablation subsection. It compares the full W-Net trained with rotational augmentation against an otherwise identical 3D U-Net baseline without augmentation, reporting improved detection metrics across a range of asteroid velocities. We also compare Adaptive Normalization against standard min-max and z-score normalization, demonstrating faster convergence and higher final accuracy. These new results are presented with quantitative tables and are now referenced in the abstract and conclusions. revision: yes

standing simulated objections not resolved

The absence of any quantitative validation or transfer experiments on actual TESS observations (as opposed to simulations).

Circularity Check

0 steps flagged

No circularity: supervised training on augmented simulations is self-contained

full rationale

The paper defines a W-Net (stacked 3D U-Nets) and Adaptive Normalization as new components, generates training cubes via the released tess-asteroid-ml simulator, and applies explicit rotation augmentation to achieve trajectory robustness. These steps are forward design choices in a standard supervised pipeline; the network output is not equivalent to the input masks or rotations by construction, nor does any claim reduce to a fitted parameter renamed as prediction. No self-citations appear as load-bearing justifications for uniqueness theorems or ansatzes, and the generalization claim to real TESS data is presented as an empirical transfer task rather than a definitional identity. The derivation chain therefore remains independent of its own outputs.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 2 invented entities

The approach depends on the assumption that deep learning models trained on simulated data can generalize to real observations, with the main innovations being the architecture and normalization technique.

free parameters (1)

W-Net model parameters
The weights and biases of the neural network are fitted during training on the augmented TESS data.

axioms (1)

domain assumption 3D U-Nets can effectively process spatio-temporal image data for segmentation tasks
This is assumed in stacking two U-Nets for background filtering and moving object identification.

invented entities (2)

W-Net no independent evidence
purpose: Stacked 3D U-Net architecture for asteroid detection in time-series data
Newly named and described architecture in this work.
Adaptive Normalization no independent evidence
purpose: Learned scaling of input data for optimal neural network performance
Novel data processing method introduced.

pith-pipeline@v0.9.0 · 5520 in / 1585 out tokens · 122017 ms · 2026-05-13T03:23:29.797903+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

178 extracted references · 178 canonical work pages · 5 internal anchors

[1]

, keywords =

KIC 2856960: the impossible triple star. , keywords =. doi:10.1093/mnras/stu1733 , archivePrefix =. 1409.0722 , primaryClass =

work page doi:10.1093/mnras/stu1733
[2]

, keywords =

Eclipse timing variation analyses of eccentric binaries with close tertiaries in the Kepler field. , keywords =. doi:10.1093/mnras/stv015 , archivePrefix =. 1412.5759 , primaryClass =

work page doi:10.1093/mnras/stv015
[3]

2016, MNRAS, 455, 4136, doi: 10.1093/mnras/stv2530

A comprehensive study of the Kepler triples via eclipse timing. , keywords =. doi:10.1093/mnras/stv2530 , archivePrefix =. 1510.08272 , primaryClass =

work page doi:10.1093/mnras/stv2530
[4]

, keywords =

The extraneous eclipses on binary light curves: KIC 5255552, KIC 10091110, and KIC 11495766. , keywords =. doi:10.1051/0004-6361/201630072 , archivePrefix =. 1710.10913 , primaryClass =

work page doi:10.1051/0004-6361/201630072
[5]

Kepler Eclipsing Binary Stars. I. Catalog and Principal Characterization of 1879 Eclipsing Binaries in the First Data Release. , keywords =. doi:10.1088/0004-6256/141/3/83 , archivePrefix =. 1006.2815 , primaryClass =

work page doi:10.1088/0004-6256/141/3/83
[6]

, keywords =

A transiting companion to the eclipsing binary KIC002856960. , keywords =. doi:10.1051/0004-6361/201219479 , archivePrefix =. 1208.3591 , primaryClass =

work page doi:10.1051/0004-6361/201219479
[7]

, keywords =

The Triply Eclipsing Hierarchical Triple Star KIC002856960. , keywords =. doi:10.1088/0004-637X/763/2/74 , archivePrefix =. 1211.2262 , primaryClass =

work page doi:10.1088/0004-637x/763/2/74
[8]

, keywords =

Stability of planetary, single M dwarf, and binary star companions to Kepler detached eclipsing binaries and a possible five-body system. , keywords =. doi:10.1093/mnras/staa2532 , archivePrefix =. 2008.07731 , primaryClass =

work page doi:10.1093/mnras/staa2532 2008
[9]

and Eberhart, R

Kennedy, J. and Eberhart, R. , booktitle=. Particle swarm optimization , year=

work page
[10]

, keywords =

REBOUND: an open-source multi-purpose N-body code for collisional dynamics. , keywords =. doi:10.1051/0004-6361/201118085 , archivePrefix =. 1110.4876 , primaryClass =

work page doi:10.1051/0004-6361/201118085
[11]

Shapely: manipulation and analysis of geometric objects , year =

Sean Gillies and others , organization =. Shapely: manipulation and analysis of geometric objects , year =

work page
[12]

Miranda , title = "

Lester James V. Miranda , title = ". Journal of Open Source Software , year =. doi:10.21105/joss.00433 , url =

work page doi:10.21105/joss.00433
[13]

, keywords =

EPIC 220204960: A Quadruple Star System Containing Two Strongly Interacting Eclipsing Binaries. , keywords =. doi:10.1093/mnras/stx143 , archivePrefix =. 1701.05281 , primaryClass =

work page doi:10.1093/mnras/stx143
[14]

, keywords =

The Discovery of Two Quadruple Star Systems with the Second and Third Shortest Outer Periods. , keywords =. doi:10.3847/1538-4357/adcece , archivePrefix =. 2504.12239 , primaryClass =

work page doi:10.3847/1538-4357/adcece
[15]

, keywords =

Mysterious Dust-emitting Object Orbiting TIC 400799224. , keywords =. doi:10.3847/1538-3881/ac2c81 , archivePrefix =. 2110.01019 , primaryClass =

work page doi:10.3847/1538-3881/ac2c81
[16]

R., Winn, J

Journal of Astronomical Telescopes, Instruments, and Systems , month = jan, number = 1, pages =. doi:10.1117/1.JATIS.1.1.014003 , eid =

work page internal anchor Pith review doi:10.1117/1.jatis.1.1.014003
[17]

, keywords =

TIC 278825952: a triply eclipsing hierarchical triple system with the most intrinsically circular outer orbit. , keywords =. doi:10.1093/mnras/staa2762 , archivePrefix =. 2009.03378 , primaryClass =

work page doi:10.1093/mnras/staa2762 2009
[18]

, keywords =

The compact triply eclipsing triple star TIC 209409435 discovered with TESS. , keywords =. doi:10.1093/mnras/staa1817 , archivePrefix =. 2006.10449 , primaryClass =

work page doi:10.1093/mnras/staa1817 2006
[19]

, keywords =

TIC 454140642: A Compact, Coplanar, Quadruple-lined Quadruple Star System Consisting of Two Eclipsing Binaries. , keywords =. doi:10.3847/1538-4357/ac04ad , archivePrefix =. 2105.12586 , primaryClass =

work page doi:10.3847/1538-4357/ac04ad
[20]

International Conference on Learning Representations (ICLR) , year=

Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs) , author=. International Conference on Learning Representations (ICLR) , year=

work page
[21]

doi:10.17909/0cp4-2j79 , url =

TESS Calibrated Full Frame Images: All Sectors , publisher =. doi:10.17909/0cp4-2j79 , url =

work page doi:10.17909/0cp4-2j79
[22]

, keywords =

Detecting Asteroids in TESS Full-frame Images Using a Gradient-based Iterative Algorithm. , keywords =. doi:10.3847/1538-3881/adea3c , adsurl =

work page doi:10.3847/1538-3881/adea3c
[23]

Layer Normalization

Layer Normalization , author=. arXiv preprint arXiv:1607.06450 , year=

work page internal anchor Pith review Pith/arXiv arXiv
[24]

International Conference on Learning Representations (ICLR) , year=

Neural Machine Translation by Jointly Learning to Align and Translate , author=. International Conference on Learning Representations (ICLR) , year=

work page
[25]

Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages=

Effective Approaches to Attention-based Neural Machine Translation , author=. Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages=. 2015 , doi=

work page 2015
[26]

Advances in Neural Information Processing Systems (NeurIPS) , volume=

Attention Is All You Need , author=. Advances in Neural Information Processing Systems (NeurIPS) , volume=. 2017 , doi=

work page 2017
[27]

International Conference on Machine Learning (ICML) , pages=

Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift , author=. International Conference on Machine Learning (ICML) , pages=. 2015 , organization=

work page 2015
[28]

, keywords =

Prediscovery TESS Observations of Interstellar Object 3I/ATLAS. , keywords =. doi:10.3847/2041-8213/ae1f91 , archivePrefix =. 2508.02499 , primaryClass =

work page doi:10.3847/2041-8213/ae1f91 2041
[29]

, keywords =

The Near-Earth Object Surveyor Mission. , keywords =. doi:10.3847/PSJ/ad0468 , archivePrefix =. 2310.12918 , primaryClass =

work page doi:10.3847/psj/ad0468
[30]

, keywords =

TIC 168789840: A Sextuply Eclipsing Sextuple Star System. , keywords =. doi:10.3847/1538-3881/abddb5 , archivePrefix =. 2101.03433 , primaryClass =

work page doi:10.3847/1538-3881/abddb5
[31]

A., Henry, T

A Survey of Stellar Families: Multiplicity of Solar-type Stars. , keywords =. doi:10.1088/0067-0049/190/1/1 , archivePrefix =. 1007.0414 , primaryClass =

work page doi:10.1088/0067-0049/190/1/1
[32]

, keywords =

The Updated Multiple Star Catalog. , keywords =. doi:10.3847/1538-4365/aaa1a5 , archivePrefix =. 1712.04750 , primaryClass =

work page doi:10.3847/1538-4365/aaa1a5
[33]

2022 , issn =

W-Net: Dense and diagnostic semantic segmentation of subcutaneous and breast tissue in ultrasound images by incorporating ultrasound RF waveform data , journal =. 2022 , issn =. doi:https://doi.org/10.1016/j.media.2021.102326 , url =

work page doi:10.1016/j.media.2021.102326 2022
[34]

PLOS ONE , volume=

W-Net: Convolutional neural network for segmenting remote sensing images by dual path semantics , author=. PLOS ONE , volume=. 2023 , month=. doi:10.1371/journal.pone.0288311 , url=

work page doi:10.1371/journal.pone.0288311 2023
[35]

, keywords =

101 eclipsing quadruple star candidates discovered in TESS full frame images. , keywords =. doi:10.1093/mnras/stad2947 , archivePrefix =. 2309.14200 , primaryClass =

work page doi:10.1093/mnras/stad2947
[36]

Revista Mexicana de Astronomia y Astrofisica Conference Series , year = 2007, editor =

Precision Radial Velocities for the Kepler Era. Revista Mexicana de Astronomia y Astrofisica Conference Series , year = 2007, editor =

work page 2007
[37]

Journal of Global Optimization , year=

Differential Evolution – A Simple and Efficient Heuristic for global Optimization over Continuous Spaces , author=. Journal of Global Optimization , year=

work page
[38]

, keywords =

Six new compact triply eclipsing triples found with TESS. , keywords =. doi:10.1093/mnras/stac957 , archivePrefix =. 2204.02539 , primaryClass =

work page doi:10.1093/mnras/stac957
[39]

, keywords =

Triply eclipsing triple stars in the northern TESS fields: TICs 193993801, 388459317, and 52041148. , keywords =. doi:10.1093/mnras/stab3397 , archivePrefix =. 2111.10074 , primaryClass =

work page doi:10.1093/mnras/stab3397
[40]

, keywords =

Ninety-seven Eclipsing Quadruple Star Candidates Discovered in TESS Full-frame Images. , keywords =. doi:10.3847/1538-4365/ac5458 , archivePrefix =. 2202.05790 , primaryClass =

work page doi:10.3847/1538-4365/ac5458
[41]

arXiv e-prints , keywords =

Multiply eclipsing candidates from the TESS satellite. arXiv e-prints , keywords =

work page
[42]

Kepler Planet-Detection Mission: Introduction and First Results.Science2010,327, 977

Kepler Planet-Detection Mission: Introduction and First Results. Science , keywords =. doi:10.1126/science.1185402 , adsurl =

work page doi:10.1126/science.1185402
[43]

2016, ApJS, 222, 8, doi: 10.3847/0067-0049/222/1/8 Ekstr¨ om, S., Georgy, C., Eggenberger, P., et al

MESA Isochrones and Stellar Tracks (MIST) 0: Methods for the Construction of Stellar Isochrones. , keywords =. doi:10.3847/0067-0049/222/1/8 , archivePrefix =. 1601.05144 , primaryClass =

work page doi:10.3847/0067-0049/222/1/8
[44]

Mesa Isochrones and Stellar Tracks (MIST). I. Solar-scaled Models. , keywords =. doi:10.3847/0004-637X/823/2/102 , archivePrefix =. 1604.08592 , primaryClass =

work page internal anchor Pith review Pith/arXiv arXiv doi:10.3847/0004-637x/823/2/102
[45]

Nature Methods , year="2020", adsurl=

SciPy 1.0: Fundamental Algorithms for Scientific Computing in Python. Nature Methods , year="2020", adsurl=

work page 2020
[46]

, keywords =

Photodynamical analysis of the triply eclipsing hierarchical triple system EPIC 249432662. , keywords =. doi:10.1093/mnras/sty3157 , archivePrefix =. 1809.04366 , primaryClass =

work page doi:10.1093/mnras/sty3157
[47]

, keywords =

TICs 167692429 and 220397947: the first compact hierarchical triple stars discovered with TESS. , keywords =. doi:10.1093/mnras/staa495 , archivePrefix =. 2002.06902 , primaryClass =

work page doi:10.1093/mnras/staa495 2002
[48]

, keywords =

The Visual Survey Group: A Decade of Hunting Exoplanets and Unusual Stellar Events with Space-based Telescopes. , keywords =. doi:10.1088/1538-3873/ac6e06 , archivePrefix =. 2205.07832 , primaryClass =

work page doi:10.1088/1538-3873/ac6e06
[49]

, keywords =

The Revised TESS Input Catalog and Candidate Target List. , keywords =. doi:10.3847/1538-3881/ab3467 , archivePrefix =. 1905.10694 , primaryClass =

work page doi:10.3847/1538-3881/ab3467 1905
[50]

2021, Universe, 7, 369, doi: 10.3390/universe7100369

Space-Based Photometry of Binary Stars: From Voyager to TESS. Universe , keywords =. doi:10.3390/universe7100369 , archivePrefix =. 2110.03543 , primaryClass =

work page doi:10.3390/universe7100369
[51]

Galaxies , keywords =

Eclipsing Binaries in Dynamically Interacting Close, Multiple Systems. Galaxies , keywords =. doi:10.3390/galaxies10010009 , archivePrefix =. 2201.01243 , primaryClass =

work page doi:10.3390/galaxies10010009
[52]

, keywords =

BG Ind: the nearest doubly eclipsing, compact hierarchical quadruple system. , keywords =. doi:10.1093/mnras/stab621 , archivePrefix =. 2103.00925 , primaryClass =

work page doi:10.1093/mnras/stab621
[53]

, keywords =

The Random Transiter - EPIC 249706694/HD 139139. , keywords =. doi:10.1093/mnras/stz1772 , archivePrefix =. 1906.11268 , primaryClass =

work page doi:10.1093/mnras/stz1772 1906
[54]

Research Notes of the American Astronomical Society , keywords =

Calibrated Full-frame Images for the TESS Quick Look Pipeline. Research Notes of the American Astronomical Society , keywords =. doi:10.3847/2515-5172/abd63a , adsurl =

work page doi:10.3847/2515-5172/abd63a
[55]

, keywords =

eleanor: An Open-source Tool for Extracting Light Curves from the TESS Full-frame Images. , keywords =. doi:10.1088/1538-3873/ab291c , archivePrefix =. 1903.09152 , primaryClass =

work page doi:10.1088/1538-3873/ab291c 1903
[56]

Research Notes of the American Astronomical Society , keywords =

The NASA GSFC TESS Full Frame Image Light Curve Data Set. Research Notes of the American Astronomical Society , keywords =. doi:10.3847/2515-5172/ac74c4 , adsurl =

work page doi:10.3847/2515-5172/ac74c4
[57]

Universe , keywords =

Architecture of Hierarchical Stellar Systems and Their Formation. Universe , keywords =. doi:10.3390/universe7090352 , archivePrefix =. 2109.09118 , primaryClass =

work page doi:10.3390/universe7090352
[58]

, keywords =

The compact multiple system HIP 41431. , keywords =. doi:10.1093/mnras/stz1510 , archivePrefix =. 1905.12051 , primaryClass =

work page doi:10.1093/mnras/stz1510 1905
[59]

, keywords =

The CoRoT satellite in flight: description and performance. , keywords =. doi:10.1051/0004-6361/200810860 , archivePrefix =. 0901.2206 , primaryClass =

work page doi:10.1051/0004-6361/200810860
[60]

, keywords =

Secular changes in the orbits of the quadruple system VW LMi. , keywords =. doi:10.1093/mnras/staa699 , archivePrefix =. 2003.08169 , primaryClass =

work page doi:10.1093/mnras/staa699 2003
[61]

, keywords =

Parallax Systematics and Photocenter Motions of Benchmark Eclipsing Binaries in Gaia EDR3. , keywords =. doi:10.3847/2041-8213/abdaad , archivePrefix =. 2101.03425 , primaryClass =

work page doi:10.3847/2041-8213/abdaad 2041
[62]

2001, MNRAS, 322, 231, doi: 10.1046/j.1365-8711.2001.04022.x

Tidal interactions in star cluster simulations. , keywords =. doi:10.1046/j.1365-8711.2001.03974.x , adsurl =

work page doi:10.1046/j.1365-8711.2001.03974.x 2001
[63]

2012 , month = apr, journal =

PARSEC: stellar tracks and isochrones with the PAdova and TRieste Stellar Evolution Code. , keywords =. doi:10.1111/j.1365-2966.2012.21948.x , archivePrefix =. 1208.4498 , primaryClass =

work page doi:10.1111/j.1365-2966.2012.21948.x 2012
[64]

Multiple Stars Across the H-R Diagram , year = 2008, editor =

Dynamics and Stability of Triple Stars. Multiple Stars Across the H-R Diagram , year = 2008, editor =. doi:10.1007/978-3-540-74745-1\_2 , adsurl =

work page doi:10.1007/978-3-540-74745-1 2008
[65]

Keras , year =

Chollet, Fran. Keras , year =

work page
[66]

Matplotlib: A 2D graphics environment , volume =

Hunter, John D , journal =. Matplotlib: A 2D graphics environment , volume =

work page
[67]

R., Millman, K

Harris, Charles R and Millman, K Jarrod and van der Walt, St. Nature , number =. 2020 , Bdsk-Url-1 =. doi:10.1038/s41586-020-2649-2 , issn =

work page doi:10.1038/s41586-020-2649-2 2020
[68]

Data Structures for Statistical Computing in Python , year =

Wes McKinney , booktitle =. Data Structures for Statistical Computing in Python , year =

work page
[69]

and Varoquaux, G

Pedregosa, F. and Varoquaux, G. and Gramfort, A. and Michel, V. and Thirion, B. and Grisel, O. and Blondel, M. and Prettenhofer, P. and Weiss, R. and Dubourg, V. and Vanderplas, J. and Passos, A. and Cournapeau, D. and Brucher, M. and Perrot, M. and Duchesnay, E. , journal =

work page
[70]

2015 , Bdsk-Url-1 =

Mart\'. 2015 , Bdsk-Url-1 =

work page 2015
[71]

ascl , author =:1812.013 , howpublished =

work page
[72]

ascl , author =:2003.001 , howpublished =

work page 2003
[73]

ascl , author =:1905.007 , keywords =

work page 1905
[74]

P\'erez, Fernando and Granger, Brian E. , doi =. Computing in Science and Engineering , month = may, number =

work page
[75]

Research Notes of the AAS , abstract =

Hedges, Christina and Martínez-Palomera, Jorge and Tuson, Amy and Pritchard, Tyler and Hounsell, Rebekah and Schanche, Nicole and Kostov, Veselin and Giles, Daniel , title =. Research Notes of the AAS , abstract =. 2025 , month =. doi:10.3847/2515-5172/adef3a , url =

work page doi:10.3847/2515-5172/adef3a 2025
[76]

1999, XSPEC: An X-ray spectral fitting package,, Astrophysics Source Code Library, record ascl:9910.005 http://ascl.net/9910.005 Astropy Collaboration, Robitaille, T

doi:10.1051/0004-6361/201322068 , Eid =. arXiv , Author =:1307.6212 , Journal =

work page doi:10.1051/0004-6361/201322068
[77]

The Astronomical Journal , author =

The Astropy Project: Building an Open-science Project and Status of the v2.0 Core Package. , keywords =. doi:10.3847/1538-3881/aabc4f , archivePrefix =. 1801.02634 , primaryClass =

work page doi:10.3847/1538-3881/aabc4f
[78]

The Astropy Project: Sustaining and Growing a Community-oriented Open-source Project and the Latest Major Release (v5.0) of the Core Package

The Astropy Project: Sustaining and Growing a Community-oriented Open-source Project and the Latest Major Release (v5.0) of the Core Package. , keywords =. doi:10.3847/1538-4357/ac7c74 , archivePrefix =. 2206.14220 , primaryClass =

work page internal anchor Pith review Pith/arXiv arXiv doi:10.3847/1538-4357/ac7c74
[79]

keras-vis , year =

Kotikalapudi, Raghavendra and contributors , howpublished =. keras-vis , year =

work page
[80]

MPI for Python: Performance improvements and MPI-2 extensions , volume =

Lisandro Dalcin and Rodrigo Paz and Mario Storti and Jorge D'Elia , doi =. MPI for Python: Performance improvements and MPI-2 extensions , volume =. Journal of Parallel and Distributed Computing , number =. 2008 , Bdsk-Url-1 =

work page 2008

Showing first 80 references.