Capturing visual sources

Supported sources

Video and ready-made media files are used as a source for viewing and selecting frames; Capture does not re-record them as a new video file.

Recording Modes

• At regular intervals: frames are added at the selected frequency.
• Around actions: frames are saved before, during and after a keyboard or mouse action.
• Single frame: current material is transferred to Selector without lengthy recording.

Frequent recording and large sources increase the load on the processor, memory and disk. Before a long session, check the selected frequency and storage directory.

Timeline

The timeline shows frame order and the associated source. You can navigate to a frame, change the range, delete unwanted frames and select material for annotation.

Removing a frame from the current sequence may affect associated actions and annotations. Save a copy of the recording before bulk editing.

Recording actions and Action CV

Action recording associates keyboard and mouse events with frames. The experimental Action CV can suggest a visual action target using an LLM for planning and local computer-vision operations for processing.

The proposed target requires user review and is saved alongside the original recording. Analysis does not replace the original actions and frames.

Network streams and credentials

When selecting a network stream, Screph connects directly to the specified address. Don't put long-lived passwords and tokens in URLs unless necessary. If the address contains access data, consider the local settings to be sensitive.

Export frames

Selected frames are exported as PNG with source and timing information. Export creates separate files and does not replace saving the project. Cancellation may leave files already written in the folder; the final manifest confirms completion of the whole export.

Before annotation

1. Check the source, scale and desired time range.
2. Delete frames you do not need, keeping a backup of the recording.
3. Select a frame and transfer it to Selector.

Source, recording, and stopping

1. On the Capture tab, select a source and its settings: a window or monitor, a screen region, a camera, a stream address, or a file. Check the source image.
2. Set the recording rate, storage method, and frame mode. After recording starts, the source appears on the timeline. Pausing playback does not stop recording.
3. Stop recording and select a frame, range, and sampling step. Send a frame to Selector or use the range for export, a dataset, or sequence processing. Save material held in memory to files for longer-term use.

A frame sent to Selector belongs to a specific source state. Switching the camera, file, or timeline does not transfer old results to new frames. Select the required frame or range again for further annotation, training, or sequence processing.

Camera calibration

1. Select a lens model and calibration target. Enter the number of inner corners for a checkerboard or the number of squares for ChArUco. Measure the physical square size. Create target… saves a printable target; print it without scaling to fit.
2. Confirm that you are using one camera with unchanged resolution, focus, and zoom, without prior lens correction. Add files, Timeline frames…, or frames from a running source. Auto-collect takes fresh, stable frames; vary the target position and tilt during collection.
3. In the table, check the detected target and each frame’s role: calculation, validation, or skip. Calculate requires at least eight calculation frames and three held-out validation frames. Duplicates, frames with no detected target, and frames of incompatible size cannot be used for calculation.
4. After calculation, review fitting and validation errors separately, frame coverage, and warnings. Stopping collection or cancelling calculation retains the accepted dataset until the application closes. After restoring a ZIP, create a new dataset for live collection; the reconnected camera is not assumed to be identical automatically.

Camera calibration… is in the Source group of the Capture tab for camera, stream, and file sources in General, Industrial, and UAV modes. This route is unavailable for screen capture and GUI Automation. The window is nonmodal; Detailed controls reveals settings and reports for the same dataset.

The JSON profile stores the calculated parameters, the ZIP dataset lets work with the frames continue later, and Add to project library saves a version in the project. Saving Selector does not replace the calibration ZIP. Correct image… processes an image from the same camera at the same size; the result can be saved as a separate PNG.

Comparing lens models requires new independent validation frames afterwards. Validate with new frames checks saved parameters without fitting them again. Calibration alone does not establish measurement accuracy on hardware or the metric scale of a monocular trajectory.