Waymo Open Dataset Versions Save

Fixes for the WOSAC metrics:

• Fixing a bug in validity checking for collision and offroad.
• Modifying the behavior of collision/offroad checking when invalid.

Installation:

To install the updated version of the pip package:

pip install waymo-open-dataset-tf-2-11-0==1.6.1

New features:

• Support object asset dataset
• A codec to encode/decode object assets
• Expose and register world_to_image_moving_point method for the camera model
• Support for different types of filtered aligned types for AlignedPredictionBox in metrics

Fixes:

• Fix the camera_segmentation_metrics test
• Fix camera_model_ops_test
• Fix a bug in the Component.schema to handle python type aliases correctly
• Fix some of pytype annotations
• Clean up *.ipynb files.
• Use rules_python blaze rules for py_binary targets

Installation:

To install the updated version of the pip package:

pip install waymo-open-dataset-tf-2-11-0==1.6.0

Fixes:

• Resolved the Pose Estimation metric issues and added a detailed description of the metric.
• Updated tutorial_keypoints.ipynb to specify authors as a list of strings.
• Fixed a bug in the 2D Semseg Tutorial.
• Corrected a bug in LET matching that occurred when the prediction and ground truth were on different sides of the sensor.
• Updated the motion tutorial with the correct number of map samples.
• Fixed issues with v2 lidar utils.
• Updated pvps test frames.
• Improved error messages in Sim Agent evaluation code.

Installation:

To install the updated version of the pip package:

pip install waymo-open-dataset-tf-2-11-0==1.5.2

This minor update includes:

Fixes:

• Fix bazel build issue under root user in colab in the tutorial.ipynb
• Specify the latest version of of the pip package in all tutorials
• Add missing build targets for wod_detection_evaluator
• Modify WOSAC evaluation to make it invariant to object ID permutation order
• Remove outdated quick_start.md

Features:

• Update camera_segmentation utils and tutorial to support the v2 dataset.
• Script to install deeplab2 dependency for camera_segmentation
• A bucketed version of the sim agents metrics, shown on the leaderboard.

To install the corresponding pip package:

pip install waymo-open-dataset-tf-2-11-0==1.5.1

This major update includes:

• Introduced a new modular dataset format based on the Apache Parquet file format (tutorial)
• For the 2D video panoptic segmentation labels, added a mask to indicate the number of cameras covering each pixel.
• Added 3D map data as polylines or polygons.
• Added Lidar data for the training set (tutorial)
• Added driveway entrances to the map data.
• Added supporting code for the 2023 Waymo Open Dataset Challenges
• Replaced old TensorFlow 2.6 build configuration with the new simplified TensorFlow 2.11 build configuration
• Reorganized directory structure: renamed third_party into wdl_limited, moved all code into src folder

To install the new package:

pip install waymo-open-dataset-tf-2.11.0==1.5.0