As already mentioned in a previous news Philosys is releasing a new label product that isn't based on EBAssist/ADTF anymore.
This new product ist the Philosys Ground Truth Annotator 2018.
Philosys Label Editor will still support ADTF for those users that would continue to use the advantages of ADTF during development and validation of ADAS/AD systems, and the seemless integration of ADTF within Philosys Label Editor. Both label tools support the same features regarding labeling.
The main new features are the support of new data formats and a much faster data player implementation than the ADTF harddisk player used up to now.
Supported Data Formats
The main issue when starting an annotation project is to provide recorded data in a format that the label tools can handle. Up to now Philosys Label Editor was tailored to the EBAssist/ADTF DAT format. When the data was already recorded in this format then Philosys Label Editor can read this format directly by using the ADTF harddisk player. If necessary, the data samples themselves can be converted to the supported data format by using ADTF filters.
But data that isn't recorded in DAT format had to be converted before use. This costs a lot of additional resources. With Philosys Label Editor V6 the new custom player interface has been introduced to alleviate this problem, by being able to implement a player for a specific data format and replacing the ADTF harddisk player, but still at the cost of a EBAssist Runtime license per user.
The first version of Philosys Ground Truth Annotator 2018 supports several alternative input data formats without having to pay extra license fees. Over the time there will be more and more commonly known formats available. Customer specific formats can be implemented separetely on customer request. Supporting recording formats directly makes handling large amounts of recordings much easier and provides considerable cost savings.
The players provide the data samples of the supported file formats as is to Philosys Ground Truth Annotator 2018. So there may be conversions necessary to use this data. Unlike Philosys Label Editor there is no filtergraph available, which allows flexible preprocessing of data between file reading and data visualization. But some of this functionality can be handled by new decoder plugins, explained later.
Philosys Sample File Format
This new file format was introduced to allow a simple setup of data for labeling. Samples have to be provided within a project directory containing a directory, with streams represented by subdirectories, containing single files for each sample. Timestamps in microseconds are provided by the sample file names of the designated master stream.
This format is generic and can represent all necessary information for labeling. It is easy for customers to start with, by exporting their samples from other formats, and placing the sample files in the directories for each stream.
Philosys Ground Truth Annotator 2018 comes with the Sample File Player-plugin as standard and supports the following sample formats delivered by a data stream:
- Video (BMP, PNG, JPEG)
- Point Cloud Points as array of floats (XYZ, XYZI, XYZRGB, ...)
- Odometry as character string defining 4x4 matrice
- External XML data as object list in Label XML format
- Binary data blobs
EB Assist DAT Format
For reading DAT format files the new StreamingLib Player-plugin is provided. It is based on StreamingLib SDK available from Elektrobit.
- Video as ADTF image format - uncompressed/compressed data type
- Point Cloud as array of float32 (XYZ, XYZI, …) - structured data type)
- Odometry as character string representing a 4x4 matrice - structured data type
- External XML data as object list using label XML format - structured data type
- Binary data blobs – structured data type
- Info strings attached to samples within master data stream
Time is provided by the sample times of the configured master stream.
The samples are stored in a DAT file that can either be created by using EB Assist, or by having a program by using the StreamingLib SDK. Compressed, splitted DAT files and ADTFv3 files are not supported.
ROS Bag Format
Philosys Ground Truth Annotator 2018 supports reading ROS bag files. At the moment only a limited number of standard topics is supported:
- Video as sensor_msgs/Image or sensor_msgs/CompressedImage.h
- Point Cloud Points as sensor_msgs/PointCloud2 - little-endian, XYZ or XYZI, value type float32 (7)
- Odometry as tf2_msgs/TF_message
Other topic types can be implemented on customer request.
Time is provided by the message time stamps of the configured master stream.
Intempora RTMaps Datasets
Philosys Ground Truth Annotator 2018 supports reading Intempora RTMaps Datasets. The following data formats are supported:
- Video as jseq, jpeg, pngseq, png and raw
- Point Cloud Points as numeric_binary with float32
- Odometry as text
Other data formats can be implemented on customer request.
Time is provided by the message time stamps of the configured master stream.
Label Editor Decoder Plugins
As Philosys Ground Truth Annotator 2018 is without ADTF filtergraph, no pre-processing of data with filters is possible. In general this shouldn't be a problem, because data could be pre-generated for direct interpretation by Philosys Ground Truth Annotator 2018. But, sometimes it is more convenient to have the possibility for pre-processing data within the label tools, i.e. when using compressed data. In this case this not only reduces the data size on storage, but also increases data rate.
To support this type of processing Philosys introduced the concept of Decoder Plugins. This plugins are acting like filters in ADTF, but have just one input and one output. Multiple plugins can be stacked on each stream to allow processing of file data. There are Decoder Plugins that do player specific processing, and others can be applied to any stream.
The following generic Decoder Plugins are currently supported:
- BMP Decoder
- JPG Decoder
- PNG Decoder
- YUV422 Decoder (binary blob)
- DAT ADTF Video Compressed (StreamingLib player only)
Because the interfaces for players and decoder plugins are not stabel yet, by now they are not available as SDK. But Philosys will implement required plugins on customer request.
All the new plugins can be licensed for both label tools with the Player Toolbox.
Plugin Interfaces
As most of the plugin interfaces in Label Editor are implemented by use of ADTF SDK, there have been new plugin interfaces be implemented. All customer plugins have to be migrated to the new interfaces for use with Philosys Ground Truth Annotator 2018. Please contact us for help.
The ADTF-specific Mixin-Interface had to be dropped. A replacement will be provided in the future.
Licensing
Whereas Philosys Label Editor was focused at large-scale EB Assist/ADTF users, most of the time with EB Assist company group licenses, Philosys Ground Truth Annotator 2018 is targed to a much broader market. Philosys Ground Truth Annotator 2018 is useful for small to large companies providing labeling service independent from specific data formats. Labeling plattforms also move more and more into the cloud for worldwide use of resources, requiring different licensing.
To cover this new requirements the licensing model for Philosys Ground Truth Annotator 2018 will change to a per user/month scheme.
Philosys Ground Truth Annotator 2018 is available in the following variants:
- AI - support for Semantic Segmentation labeling only
- 2D - support for 2D image labeling
- 3D - support for 2D image labeling and 3D point cloud labeling
There are further license options available. Please contact us for details at This email address is being protected from spambots. You need JavaScript enabled to view it.
Please don't hesitate to contact us, if you have specific requirements for Philosys - Ground Truth Annotator 2018 or if you want to be part of our Beta program.
The Philosys Label Editor and also the Philosys Ground Truth Annotator are used during development and test of diverse Advanced Driver Assistance Systems (ADAS) for ground-truth-data annotation. Objects are manually or automatically marked and tagged with detailed traits. All kinds of vehicles, lane boundaries, traffic signs as well as pedestrians and wildlife animals are registered and verified by the assistance system for use during validation, and the data can be used for generating reference data for deep neural network (DNN) maschine learning and validation.
The ADTF based Philosys Label Editor is available as of the beginning of 2011, and the ADTF-free Philosys Ground Truth Annotator is available since mid-2018. Their many features facilitate the annotation of video scenes and reference data. This results in a significant cost reduction for the annotation process.