Find sources that change between your frames.

Transient Finder looks for things that stay in one place on the sky but change brightness: a star that flares, a source that appears where nothing was before, or one that fades away. You give it several images of the same field taken at different times. It finds point sources in each frame, groups detections that share a sky position, measures each position on every frame, and keeps the positions whose brightness changed more than static stars and seeing would explain.

It does not look for moving objects. For asteroids and comets, use Asteroid/Comet Detection.

• Search a sequence. Open a folder of repeated frames and search them in one run. Step 04
• Solve missing WCS. Frames without a usable plate solution are solved automatically and cached. R4
• Compare against one Gaia catalog. One Gaia DR3 query (G ≤ 18) covers the whole sequence and gives each candidate its nearest catalogued star. R4
• Read a ranked candidate list. Candidates are numbered by how strongly they varied, with a report in the Work Log. Step 06
• Inspect each candidate. See per-frame detections, variability, and flux ratio, with crosshairs on the image. Step 07
• Blink it. Cycle through every frame at the candidate's position to judge it by eye. Step 08
• Check it. A short checklist for deciding what a candidate is before you trust it. Step 09
• Export the blink. Save the blink as a GIF or MP4. Step 10
• Optional model scores. A local model can score candidates from saved training labels. R6

Images

At least two images of the same field taken at different times. FITS (.fit, .fits), XISF, TIFF, and camera RAW files are searched. JPG and PNG files are skipped. Similar pointing, filter, and exposure give cleaner results.

Folder layout

Put the frames directly in one folder, or in subfolders one level below it. The exact rule is in R4.

Plate solving

Frames with a valid embedded WCS are used as they are. Unsolved frames are solved during the search. For the most reliable solving, add an astrometry.net key in Settings > Open Settings > General > Astrometry API Key.

Timestamps

Observation times come from DATE-OBS. Times without a timezone are read in Image Timestamp Timezone (Settings > Open Settings > Differential Photometry, under Science Export; default UTC).

Internet

The Gaia query and astrometry.net need a connection. Solved frames and catalog results are cached and reused on the next search.

Color data

Color images, including demosaiced camera RAW, are averaged into one synthetic luminance plane. That is fine for finding changes; for precise measurements use calibrated single-channel frames.

1. On the mode picker, under SCIENCE WORKFLOWS, click Transient Finder.
2. If CAst is already open in another mode, choose Mode > Transient Finder instead.

The workflow row along the top shows Open, Train Model, Min Frames, Threshold, and ROI. Below it, the candidate table is on the left with the Work Log and Inspector tabs underneath, and the image view is on the right.

1. Click Open on the workflow row. You can also choose File > Open Folder.
2. In Select transient image folder, choose the folder that holds your frames and confirm.

CAst clears any earlier results, writes Opened transient image folder: … to the Work Log, and shows the first image it finds in the image view. The Open button changes to Search. The search has not started yet.

• Ctrl + Shift + O: File > Open Folder

1. Check Min Frames (default 2), Threshold (default 5.0 sigma), and ROI (default 3%). The defaults are a good first run. What each one does is in R1.
2. Click Search.
3. Watch the Work Log. The button reads Searching..., the three settings are locked, and a busy bar runs under the workflow row.

The search runs in this order, and each stage writes timestamped lines to the Work Log:

1. WCS check. [WCS 3/12] Checking … for each frame. Frames without a usable WCS are solved; frames that cannot be solved are skipped.
2. Gaia query. One Gaia DR3 query for the combined footprint of all solved frames.
3. Detection. [Transient 3/12] Detecting point sources in …, then the number of detections kept in that frame.
4. Screening. Detections are grouped by sky position, every group is measured on every frame, and groups that did not change enough are rejected.

When it finishes, the report is added to the Work Log, the table fills, and the first candidate is selected. The frames load in the background so blinking starts quickly.

The report

Scroll to the end of the Work Log. The report lists:

• Transient Finder scanned N image(s).
• Solved frames: N/M (K solved through astrometry.net).
• Gaia comparison sources loaded: N.
• Point-source detections retained before variability screening: N.
• Variable transient candidates retained: N.
• Top candidates: a summary line for up to 12 candidates.
• Notes: skipped frames, how many groups were rejected as static or seeing-driven, Gaia problems, and caps.

Check the solved-frame count first. If frames were skipped, the notes say why.

The candidate table

Each row is one sky position, named TF-001, TF-002, and so on. TF-001 changed the most relative to its noise. The columns are Candidate, RA, Dec, Frames, Median SNR, Nearest Gaia, First UTC, Last UTC, Label, and ML Score; their exact meaning is in R2.

1. Click a column header to sort by that column. Click it again to reverse the order.
2. Click a row to select that candidate.

1. Click a candidate row. A crosshair marks it on the image.
2. Click Center Candidate to turn it on. From now on, every candidate you select is centered in the view, at your current zoom.
3. Scroll over the image to zoom in around the cursor. Drag with the left mouse button to pan.
4. Click the Inspector tab to read the candidate's numbers.

The Inspector shows the summary line (for example variable by 41.2 sigma (flux ratio 3.4); seen above threshold in 3/5 frame(s)), then Detections, Max SNR, Variability SNR, Flux ratio, the nearest Gaia source, and Per-frame detections: with the file, time, pixel position, RA, Dec, and SNR for each frame where the source was detected.

Adjust the display

1. Click Display above the image.
2. Choose a Stretch: (Auto Stretch, Linear, Asinh, Sqrt, or Log).
3. Click Curves to shape the tone curve, then click Apply.
4. Tick Invert to show dark stars on a light sky. Faint changes are often easier to see this way.
5. Click Reset to return to the default display.

Look at several candidates together

Ctrl-click or Shift-click rows to select several candidates. All of them are marked on the image: the current one in amber, the others in blue. The Inspector, Center Candidate, Blink, and Export Blink use the current one.

• Scroll: zoom around the cursor
• Left-drag: pan
• Ctrl + click, Shift + click: select several rows
• Ctrl + A in the table: select all candidates

1. Select a candidate.
2. Choose a blink interval in the list next to Blink: 0.10 s, 0.18 s, 0.35 s (default), or 0.70 s.
3. Click Blink. A Preparing Blink window loads the frames, then playback starts with the candidate centered.
4. Watch the crosshair. Click Blink again to stop.

The blink runs through every frame where the position could be measured, in time order, including frames where nothing was detected. That is how a source that appears or disappears becomes visible: you see the empty frames as well as the bright ones.

A candidate has passed a numeric test. It may still be an artifact, a known object, or a bad frame. Work through these checks before you treat it as interesting.

1. Blink it and look closely. A real source looks like a star where it is bright, at the crosshair. A single sharp pixel is usually a hot pixel or cosmic ray. A streak is a satellite or aircraft. A dot that shifts between frames is a moving object.
2. Look at the faint frames. If the faint frame is cloudy, hazy, or out of focus, many stars get flagged together. Sort by First UTC or Last UTC and see whether a large group of candidates shares the same frames.
3. Read Nearest Gaia. A Gaia star a few arcseconds away usually means the candidate is that star. It may be a known variable, or the change may come from saturation or seeing.
4. Check the catalogs in Sky Explorer. Open one of the frames in Sky Explorer and run Explore with variable stars and asteroids/comets included. If a catalogued object sits at the position, the change is probably already known. Comparing the field with a Survey image shows whether the source was there before your frames.
5. Check for motion. If the source drifts from frame to frame, search the same folder in Asteroid/Comet Detection.
6. Measure it. If it survives these checks, measure a light curve in Differential Photometry. Transient Finder reports instrumental flux ratios and signal-to-noise ratios, not magnitudes.

1. Select the candidate, and set the zoom, stretch, and blink interval you want.
2. Click Hide Info if you do not want the information panel in the animation.
3. Click Export Blink.
4. In Export Transient Blink Animation, choose Animated GIF Files (*.gif) or MP4 Video Files (*.mp4). The default name is TF-xxx_transient_blink.gif in the opened folder.
5. Click Save. An Export Blink progress window steps through the frames; click Cancel there to stop.

Each frame of the animation is the image view exactly as you see it: its size, zoom, stretch, curves, invert, crosshairs, and the information panel if shown. Each frame lasts one blink interval, and GIFs loop forever. The Work Log confirms the saved path.

The blink animation is the one file export in this mode. The Work Log text, including the report, can be selected and copied.

Min Frames

The search needs at least this many solved frames, and a candidate must be measurable in at least this many. It does not require that many detections: a source seen in one frame can pass if enough frames could be measured at its position. The variability test always needs at least two measured frames, even at 1.

Default 2, range 1–99.

Threshold

Detection threshold in units of the frame's background noise. Raising it finds fewer, brighter sources and also raises the variability requirement, the “absent” limit, and the galaxy-core exception (see R5). Lowering it finds fainter sources and more false alarms.

Default 5.0 sigma, range 1.0–50.0 in steps of 0.5.

ROI

Border excluded from detection, as a percentage of the shorter image side. The excluded border is drawn on the image as you change the value. Even at 0%, sources within 6 px of the edge are not detected, and positions within 17 px of the edge cannot be measured.

Default 3%, range 0–45%.

Fixed values

Not adjustable in the panel: detection FWHM 3.0 px, grouping radius 2.5 arcsec, Gaia limit G ≤ 18.0, Gaia neighbor radius 5 arcsec, up to 25,000 detections per frame, and up to 500 candidates.

Related settings

In Settings > Open Settings > General: Astrometry API Key, Astrometry Timeout (default 300 s), Image Display (default stretch and invert), and Keep mode memory when switching. In Settings > Open Settings > Differential Photometry: Image Timestamp Timezone.

Sorting by a column header is the way to reorder the table. There is no filter box and no right-click menu.

Lower tabs

Work Log

Timestamped progress, the search report, and blink and export messages.

Inspector

The selected candidate's summary, Detections, Max SNR, Variability SNR, Flux ratio, nearest Gaia source, any label or model score, and the per-frame detection list. Frames where the source was measured but not detected are not listed.

Panel sizes

Drag the splitters to resize the table, the tabs, and the image. If you collapse the tabs, click ▲  Work Log · Inspector to restore them. If you collapse the left side, click ▶  Candidate Transients.

Display

Opens Stretch: (Auto Stretch, Linear, Asinh, Sqrt, Log), Curves, Invert, and Reset.

The defaults come from Image Display in Settings.

Curves

Opens the Curves dialog with a preview and histogram. Apply keeps the curve, Cancel discards it, and Reset straightens it.

Reset

Restores the default stretch and invert and removes curves.

Center Candidate

Toggle. When on, selecting a candidate centers it at the current zoom.

Blink interval

0.10 s, 0.18 s, 0.35 s, or 0.70 s per frame.

Default 0.35 s. Also sets the frame duration of exported animations.

Blink

Toggle. Cycles through every measured frame of the current candidate in time order, keeping it centered. Needs at least two frames.

Export Blink

Saves the blink as GIF or MP4. Needs at least two frames.

Hide Info / Show Info

Hides or shows the information panel on the image, which lists the file and the selected candidate's details.

Crosshairs

Amber for the current candidate, blue for other selected candidates. On a frame with no detection, the crosshair is placed from the candidate's RA and Dec through that frame's WCS.

Mouse

Scroll to zoom around the cursor; left-drag to pan. Right-click does nothing in this mode.

Which files are searched

The scan does not go through every subfolder. CAst uses the first of these that contains images:

1. If the folder has a Files subfolder: the images inside each subfolder of Files.
2. Otherwise: the images inside each direct subfolder of the opened folder.
3. Otherwise: the images directly in the opened folder.

If the folder has loose images and also subfolders with images, the loose images are not searched. All searched images form one sequence, sorted by DATE-OBS and then file name.

Plate solving

Each frame's embedded WCS is checked first. If it is missing or unusable, two solvers start together and the first success is used:

Solved frames are saved in the cache directory under transient-wcs and reused. A frame that cannot be solved is skipped with a note in the report. File > Clear All Cache… deletes all cached solutions and catalogs, keeping your settings.

The shared Gaia catalog

After solving, CAst builds one footprint that covers every solved frame and makes one Gaia DR3 query down to G 18.0, in tiles when the field is wide. It is cached under catalogs. The report line Gaia comparison sources loaded gives the count. This catalog supplies each candidate's Nearest Gaia and the rejection of ordinary stars described in R5. If the query fails, the search continues without it and the notes say Gaia lookup unavailable for this field.

Detection

Each solved frame is reduced to one plane, its background median and noise σ are estimated with sigma clipping, and a star finder (FWHM 3.0 px) detects sources above Threshold × σ. Sources inside the edge margin (the larger of 6 px and the ROI) are dropped. Detection SNR is peak / σ.

Grouping

Detections from all frames that lie within 2.5 arcsec of each other are linked into one group; each frame contributes at most one detection per group. The group's position is the mean of its detections.

Measuring every frame

Each group's position is projected into every solved frame and measured with an aperture of 5.25 px radius and a background annulus from 9.0 to 14.25 px. Flux F is the aperture sum minus the annulus background, its uncertainty is σ_local × √N_aperture, and aperture SNR is F / σ_F. Frames where the position is within 17 px of an edge are not measured.

The variability test

CAst compares the measured frame with the highest flux (bright) and the one with the lowest (faint):

• variability SNR = (F_bright − F_faint) / √(σ_bright² + σ_faint²)
• flux ratio = (F_bright + floor) / max(F_faint, floor), where floor = max(σ_bright, σ_faint, 1)

A group becomes a candidate when all of these hold:

1. At least two frames, and at least Min Frames, were measured.
2. The bright frame has positive flux with aperture SNR at least Threshold.
3. Variability SNR is at least the larger of 7.0 and 1.25 × Threshold (8.75 at the default).
4. And one of: the flux ratio is at least 2.0; the faint frame is absent (flux ≤ 0 or aperture SNR ≤ the larger of 2 and 0.55 × Threshold); or the galaxy-core exception applies (variability SNR at least the larger of 28 and 5 × Threshold, flux ratio at least 1.35, and the highest peak SNR at least 1.45 times the lowest).

The galaxy-core exception lets a strong brightening on a bright galaxy nucleus pass even when the total flux in the aperture changes by less than a factor of two.

Gaia neighbor rejection

If a Gaia star (G ≤ 18) lies within 5 arcsec of any of the group's detections and the flux ratio is below 2.0, the group is rejected. This removes ordinary stars that flicker with seeing. It applies even when the absent or galaxy-core test passed, so a real event next to a Gaia star needs a flux ratio of 2.0 or more. The report counts every rejected group as static or seeing-driven.

Ranking

Candidates are sorted by variability SNR, then flux ratio, then maximum detection SNR, and numbered in that order. At most 500 are kept.

Transient Finder can score candidates with a local Random Forest model trained on candidates you have labeled before. Labels are Real, Artifact, Known Object, Moving Object, Noise, and Unsure. Labels and models are stored on your computer in candidate-training.sqlite3.

Train Model

Trains a new model from all saved transient labels and re-scores the table. The result appears in the Work Log.

Needs at least two labeled candidates with two different labels. With six or more labels and at least two per label, a held-out validation runs as well.

Label column

The label saved for this candidate, or -.

ML Score column

The latest model's predicted label and its confidence, or - when no model exists.