saadtaleb/tess-lightcurves-planets

TESS Light Curves for Planet Detection

A curated set of TESS light curves with labels for training transit detection models. Each star in this dataset either has a reported planet candidate, a confirmed planet, a false positive, or a pipeline detection from the public NASA/TESS catalogs.

What's inside

 7 607  stars (TIC targets)
13 254  catalog signals (TOI / CTOI / TCE)
31 345  light curve FITS files

This dataset mixes three different things. Keep them separate in your head:

• Star (TIC) — a physical star in the sky, identified by its TESS Input Catalog ID.
• Light curve (FITS file) — one observation of that star by TESS: a time series of brightness vs. time. One star is usually observed several times (in different TESS "sectors" of ~27 days each), so one star can have many FITS files.
• Signal — an announcement from a public catalog that says "there's a transit around this star, here are its period / depth / duration". Signals come from three catalogs:
  • TOI — TESS Objects of Interest (official TESS team)
  • CTOI — Community TOI (community contributions)
  • TCE — Threshold Crossing Events (NASA automatic pipeline)

A star can have several signals (several planets, or the same planet listed in several catalogs), and each signal can be labeled as candidate, confirmed, false_positive, or unknown.

Concrete example

TIC 279741379  ← one star
├── 2 signals from the TOI catalog:
│   ├── TOI 186.01  (period 35.6 d, confirmed planet)
│   └── TOI 186.02  (period  7.8 d, confirmed planet)
└── 5 light curve FITS files (observations by TESS):
    ├── sector  1 — standard cadence (lc)
    ├── sector  2 — standard cadence (lc)
    ├── sector  3 — standard cadence (lc)
    ├── sector  4 — standard cadence (lc)
    └── sector 28 — fast cadence    (fast-lc)

One star. Two known planets. Five observations.

File layout

Light curves are stored by TIC ID:

TIC_<tic_id>/<filename>.fits

Three FITS product types are present:

*_lc.fits       standard TESS/SPOC 2-minute cadence
*_llc.fits      QLP pipeline (FFI-based)
*_fast-lc.fits  fast-cadence (20-second)

The 3 metadata CSVs

1. targets.csv — 1 row per star (7 607 rows)

One line per TIC. Use this when you want "what stars are in the dataset".

Columns: tic_id, system_class, stellar properties (ra_deg, dec_deg, tess_mag, stellar_eff_temp_k, stellar_radius_r_sun, …), and counts (n_signals_total, n_fits_total, etc.).

system_class tells you at a glance what's around the star:

confirmed_only  star has at least one confirmed planet
candidate_only  star has only candidate-like signals
mixed           both confirmed and candidate-like signals
unknown         star is in the dataset but no clear label was matched

2. signals.csv — 1 row per catalog signal (13 254 rows)

One line per announced transit. Use this when you want ground-truth labels.

Columns include: signal_uid, tic_id, signal_source (TOI/CTOI/TCE), signal_id, signal_label, tfopwg_disposition, period_days, epoch_bjd, duration_hours, depth_ppm, planet_radius_r_earth, training_ready_minimum.

signal_label values:

candidate       a candidate transit
confirmed       a confirmed planet
false_positive  rejected / not a real planet
unknown         no clear disposition

3. lightcurves.csv — 1 row per FITS file (31 345 rows)

One line per observation file. Use this when you want "for this file on disk, what's the star, what's known about it, and where is the file".

Columns: hf_path (path inside this HF repo), file_name, tic_id, sector, file_type, pipeline_guess, system_class, n_signals, signal_uids, signal_labels, signal_sources.

signal_uids / signal_labels are semicolon-separated lists because a star can have several signals — every signal from the same star applies to every FITS file of that star.

Label distribution

Stars:

Stars with at least one confirmed planet   1 181   (753 confirmed_only + 428 mixed)
Stars with only candidate-like signals     5 808   (candidate_only)
Stars with mixed labels                      428   (mixed — also counted above)
Stars with no clear label                    618   (unknown)

Signals (a single star can have several):

candidate        9 895
unknown          1 381
confirmed        1 312
false_positive     666

The mix is intentional: to train a transit detector, you need confirmed positives, clear negatives, and unlabeled / ambiguous cases.

Data sources

• MAST / TESS — light curve FITS files.
• ExoFOP-TESS — TOI and CTOI catalogs.
• MAST / STScI — TCE catalog.
• NASA Exoplanet Archive — confirmed TESS planets.

All upstream data are public archives. NASA's open-data policy defaults to Creative Commons Zero (CC0) for unrestricted NASA mission data. Please cite the upstream archives (TESS mission, MAST/STScI, ExoFOP-TESS, NASA Exoplanet Archive) when using this dataset in scientific work.

Citation

If you use this dataset, please cite the upstream providers:

• TESS mission (Ricker et al. 2015)
• MAST / STScI
• ExoFOP-TESS
• NASA Exoplanet Archive
• lightkurve (Lightkurve Collaboration 2018)