Please enable JavaScript.
Coggle requires JavaScript to display documents.
KPI Alignment - Coggle Diagram
KPI Alignment
Features
Pandas Dataframes in Akash's format
required by perception team
required by AICore
Data
Single Log KPI's
Support computation for log segments (timestamp A to B)
Multi-Log KPI's
Average precision
ROC curves
Raw Data
Annotations
Algorithm Results
No hard disk dependencies (writings files to disk as mandatory requirement)
KPI Summaries
Limit KPI computation to overlapping algorithm and annotation segments
Support Auto-GT features
occlusion
oncoming
Support distance dependent KPI evaluation
range specific
support list of range bins
Example
ROC range blanket
z axis -> precision
x axis -> recall
y axis -> range bins
Support classdependent KPI evaluation
Ground Truth can be
manual annotations
algorithm streams
Support angle dependent KPI evaluation
in one go support multiple angle ranges at once
Option to interpolate or correct detection streams
Computational concept layers
1) Sub-Select data on log level (only relevant data)
3) Matching with ground truth
2) Class mapping
4) Chopping into subsets
range
class
...
confidence bins
Supports ROC curves
flags
5) Compute KPI on subset
6) Provide pandas dataframe and summary
Interfaces
Microservice compatibility
Available library imports to get pandas dataframe
Configuration via
dictionaries
pip installable package
versioning
Class mapping at runtime by
dictionaries
KPI library documentation
Nexus Integration
KPI streams