What’s New

v1.0.0 (unreleased)

CircleDetector and CircleTracker: removed detection_method, marker_size, threshold_window_size, min_area, max_area, circularity, convexity, and inertia arguments.
VideoDisplay: removed overlay_pupil, overlay_gaze, overlay_circle_marker and overlay_circle_grid arguments.
BaseConfigurable: removed get_params and get_constructor_args methods (now private methods with the same name).
Packet: removed get method.

New get_test_recording(), load_gaze(), load_pupils(), load_markers(), load_motion(), merge_pupils() functions.
New PupilReader and MarkerReader classes.
New Calibration.batch_run(), CircleDetector.batch_run(), GazeMapper.batch_run() and PupilDetector.batch_run() methods.
CamParamEstimator: new grid_scale argument.
VideoReader: new undistort argument.
PupilDetector: new resolution and focal_length arguments.
Pipeline: can now be used as a context manager.
VideoReader.read_frames(): New raw, return_timestamp and return_index arguments

The test recording in the examples is not part of the package anymore but instead downloaded and cached through the pooch library. This means that TEST_RECORDING is not working anymore and is replaced by get_test_recording.

Updated the default test recording to contain post-hoc gaze mapped by Pupil Player v3.

Fixed exposure_mode="auto" not working in VideoDeviceUVC.
Fixed extrinsics not being saved by CamParamEstimator.
Added available_controls attribute to VideoDeviceUVC that returns legal values for UVC controls and controls can be set by assigning a mapping from control names to new values.

Package is now available via conda from our own channel.
Devices, streams and processes can now be used as context managers which starts and stops them automatically. This makes the Session context manager obsolete, which is being deprecated.
New VideoFileSyncer process for syncing video streamed from files.

Key presses from VideoDisplay windows are broadcast and stored in the keypresses deque by the manager.
CircleGridDetector now accepts a scale parameter to speed up detection at the expense of accuracy for high-resolution streams.
CamParamEstimator now shows a persistent overlay of previously detected circle patterns for the stream it is attached to.
Processes and pipelines have access to their parent stream via the context attribute, if applicable.
Processes can add display hooks to the packet that are picked up by VideoDisplay.
New load_pldata and save_pldata top-level methods.

Set default exposure mode of VideoDeviceUVC to "auto".
Added max_width parameter to VideoDisplay.
Disabled showing plots in Validation class.
Fixed loading of broken recordings (with missing timestamps or corrupt pldata files).
Fixed bug when loading 3D gaze with only binocular data.
Fixed support for configs with device_uid=None.

Loaded gaze now contains the eye data that produced the mapping (left, right, both eyes) as well as the 3D eye centers and gaze normals.
The load_raw_frame, load_frame, read_frames and load_dataset methods of VideoReader and OpticalFlowReader now accept both timestamps and indexes as parameters.
Streams can be started even when the underlying devices aren’t connected with allow_failure=True.
Added scale, detection_method and marker_size parameters to CircleDetector to allow more fine-grained control over detection. detection_method="vedb" is a new detection method that is less stable but significantly faster than the default.
Added a Validation process that extends Calibration, adding plots showing the coverage of the world camera FOV by the calibration marker and of the eye camera FOV by the pupils.
Added a new example validate.py that demonstrates usage of the new circle detection method and the validation.
Added experimental support for a PyAV-based video encoder (VideoEncoderAV).
Added nan_format parameter to StreamManager.format_status.

Fixed devices not being able to shut down when in restart loop.
Fixed blank window opening when starting a VideoDisplay process with paused=True.

Mostly feature-complete release that supports streaming, recording and processing of Pupil Core streams as well as reading of recordings.