Open mihaibujanca opened 5 years ago
Questions to be answered:
What is the appropriate structure? Perhaps something along the lines of
datasets = ["path/to/dataset","path/to/dataset2"]
algorithms = ["path/to/algorithm.so", "path/to/algorithm2.so"]
metrics = [metric1,metric2] # metrics required
[algorithm-name]
# algorithm settings, hyperparameters etc (e.g vocabulary path for orb-slam2)
param1 = "value"
param2 = "value2"
How do we share the access to the global configuration between the algorithms and the core of the framework? This could either be done by adding the GlobalConfiguration
as a public member variable in SLAMBenchConfig
since that is shared with the algorithms or there could be a singleton with global access.
We want to move away from the parameter manager type reading and use TOML to configure SLAMBench.
This will require creating a
class
orstruct
(e.g calledGlobalConfiguration
or similar) to replicate the same structure the config files might have. To get started on this issue, the current pipeline is roughlySpecial care to be taken of the callbacks created in SLAMBenchConfiguration. When parsing the TOML file for algorithms, make sure to call
Library_callback
. Similarly, check any other callbacks associated with other components or actions (e.g datasets)