1.2.1rc1

First release candidate for 1.2.1

1.2.0

Process

• Rename run_config.yml to pipeline_config.yml. Running PeekingDuck without the --config_path flag would be successful if either config file is present.
• Re-ran MacBook Pro FPS benchmarks on VM with 2.8 GHz 4-Core Intel Xeon (2020, Cascade Lake) CPU and 16GB RAM
• Add --verify_install flag for verifying PeekingDuck installation
• Add --num_iter flag to stop the pipeline after running a number of iterations

Features

• Add 6th category of node types - augment
• Add new categories of object tracking and crowd counting models
• Add augment.brightness, augment.contrast, dabble.statistics, dabble.tracking, draw.heat_map, input.visual, model.csrnet, model.fairmot, model.jde, model.movenet, model.yolox nodes
• Add option to select object detection classes by name in addition to ID
• Add obj_attrs and density_map data types

Deprecations and Removals

• Deprecate run_config.yml default config file - replaced by pipeline_config.yml
• Deprecate obj_tags and obj_groups data types - replaced by obj_attrs
• Deprecate draw.image_processor node - replaced by augment.brightness and augment.contrast
• Deprecate input.live and input.recorded nodes - replaced by input.visual

Bug Fixes

• Add missing RGB conversion to model.yolo and model.efficientdet, and re-ran accuracy benchmarks
• Fix model.mtcnn returning negative bounding box coordinates by clipping returned bboxes data type to [0,1] for all model nodes

Improved Documentation

• Update license and license headers to 2022
• Revamp Read the Docs layout for greater readability and accessibility
• Update standard and custom installation instructions
• Add tutorials, glossary for all built-in data types, FAQ and Troubleshooting page
• Add crowd counting, people counting (over time), and object counting (over time) use cases
• Categorise use cases into privacy protection, smart counting and COVID-19 prevention and control
• Add Jupyter Notebook for reference

Dependencies

• Add support for Python 3.9
• Add Pytorch torch == 1.10.0 and torchvision == 0.11.1
• Limit TensorFlow version to >= 2.2, <2.8
• Upgrade OpenCV opencv-contrib-python >= 4.5.2.54

Refactor

• Refactor draw.legend and draw.tag nodes for more flexibility
• Refactor dabble.check_large_groups, dabble.group_bbox_and_tag, draw.tag nodes to take in the obj_attrs data type
• Refactor dabble.check_nearby_objs, dabble.group_nearby_objs nodes to return the obj_attrs data type
• Simplify config names for model.mtcnn, model.yolo, model.yolo_face, model.yolo_license_plate nodes

1.2.0rc3

Third release candidate for 1.2.0.

1.2.0rc2

Second release candidate for 1.2.0

1.2.0rc1

First release candidate for 1.2.0

v1.1.1

Process

• Add option to create custom nodes via CLI, simplifying the process

Features

• Update threading feature for input.live to be used in Linux
• Add buffer_frames feature for input.live to allow usage with output.media_writer to produce smooth output video without losing video frames
• Add threading and buffer_frames features for input.recorded for potential speed-up
• Update draw.mosaic_bbox to run faster
• Add --log_level CLI parameter to allow changing of log level
• Add option to customise where model weights are to be downloaded via the weights_parent_dir configuration

Deprecations and Removals

• Remove size configuration for all YOLO models from Read the Docs as it is not changeable

Improved Documentation

• Rename misleading Python API feature to Import PeekingDuck as a Module
• Adopt collapsible navigation bar in Read the Docs to reduce clutter
• Revamp model section of Read the Docs to have separate sections for Object Detection and Pose Estimation

Dependencies

• Reduce required dependencies by only installing them if the node is used. These optional dependencies are listed in optional_requirements.txt.

Refactor

• Change Python code formatter to Black and update code
• Update coding conventions and CONTRIBUTING.md

v1.1.1rc1

First release candidate for v1.1.1

v1.1.0

Process

• Rename generated folder for storing model weights from weights to peekingduck_weights to avoid possible conflicts with other installed packages
• Update setup.cfg to only allow installation if Python version is 3.6, 3.7 or 3.8

Features

• Add use case: Privacy Protection (Faces)
• Add use case: Privacy Protection (License Plates)
• Add use case: Face Mask Detection
• Add model.mtcnn node
• Add model.yolo_face node
• Add model.yolo_license_plate node
• Add draw.mosaic_bbox node
• Add draw.blur_bbox node

Improved Documentation

• Add installation instructions for M1 Mac
• Fix sample code snippet in docs for custom nodes
• Add documentation badge linking to Read the Docs on README.md

v1.1.0rc1

First release candidate for v1.1.0

v1.0.3

Process

• Model weights and sample videos for download moved from Microsoft Azure to Google Cloud Storage

Features

• Add colours to logging output (e.g. yellow for warning, red for errors)
• Add peekingduck nodes CLI command to display available nodes and links to docs

Improved Documentation

• Replace Discord server communication channel with AI Singapore's community page
• Add draw.bbox node config show_labels in docs