v0.2.3

This release is identical to 0.2.2, but solves version mismatch bug due to CI/CD pipeline problem.

v0.2.2

Changes in flags for UI

• Removed the --ui_packaging_path flag which was slightly confusing, and instead added a new output format (--output_format ui), which makes more sense as the UI merely consumes another form of output form iCatcher+.

• Input videos are not copied to output when --output_format ui is passed, as this is redundant (frames are being produced, and if user needs an annotated video there are other options for that).

• Adds some basic tests for this option, and updates the README / docs accordingly

v0.2.1

Integration of GUI from Quest

• Users can now launch the offline web app using the --app flag, to interact with the Icatcher+ output files.
• When predicting on a video, the --ui_packaging_path "your_path" needs to be passed which creates a folder you can drop into the web app and start interacting with the annotations.
• Prebuilt app will be available by the pip installation. Otherwise, nodejs is required.

v0.2.0

Major updates to command line flags

• added the "--mirror_annotation" flag that swap the left and right classes for output
• added the "--overwrite" flag, which is off by default that allows icatcher to overwrite output annotation files if they exist. if they exist and this flag is off, icatcher+ will terminate.
• any command line option relating to the face detector has now a prefix "fd".
• the --dont_buffer flag was replaced with --fd_parallel_processing, and is OFF by default - users were experiencing crashes and RAM overload which should not be expected as this flag is trading more memory for better performance.
• the --num_cpus_saved option was replaced with --fd_num_cpus, now indicating how many cores would the user want to use (if --fd_parallel_processing is on). -1 means use all cores, which is the default.
• any command line flag that is experimental or that should not be changed is marked with a warning
• removed unused command line flags
• better tests for the full pipeline

v0.1.2

This release does not have any functional changes, but allows for a docker image to be created automatically.

v0.1.1

Second major upgrade to iCatcher+ from Quest

• Gaze prediction model defaults to regnet (on lookit) instead of resnet (on lookit)
• Better tests
• Slightly changed command line option "--model" to indicate the name of the model file to use (out of predefined choices), instead of a custom path. We will not support arbitrary model files in the official release.

v0.1.0

First major upgrade to iCatcher+ from Quest

• Face detector default is now RetinaFace, see command line options for full customization. The older opencv_dnn is still available.
• Fixes bug with passing crop_mode "right" not working correctly
• Better logging
• Better testing

v0.0.14

This will be the last release before Quest upgrades to core.

• Fixes a bug where if user specifies negative illegal transitions becoming positive, then program crashes #34
• Minor upgrade to "lowest face selector", where now the ratio of the bounding box is also being considered. This was found to improve detection in some edge cases where parents are in the frame, and shouldn't change prior behavior for any case where only an infant is present #40.
• Minor upgrade to probing for VFR using ffmpeg, where wrongly structured videos can still be probed by positive look-ahead instead of crashing #35

v0.0.13

First automated release.