In order to share code we need some data, at least for testing and demo.
The SRP data is tricky to share (I have sent an email asking permission already).
We need to find public mobile or dense LiDAR or range data with imagery to go with it.
What we need is a data folder with Makefile that pulls down the data and then processes into a form we can use.
I should be able to do this:
cd data
make kitti
make nyu
make ford
or just
make data
The result should be that the datasets are downloaded (I suggest targets like download-kitti) and then extracted (extract-kitti) and then split into batch-sized-chunks. Put the intermediate data in folders with names that uniquely depend on the way data is processed, like data/intermediate/WxHxD-0xbfff/*.pkl, where 0xbfff is supposed to be a hash of settings like the volume-min, volume-max, and any other config settings that might effect the output.
Use Makefiles so that data only gets processed when it is absolutely needed.
jfemiani
commented
6 years ago
In order to share code we need some data, at least for testing and demo. The SRP data is tricky to share (I have sent an email asking permission already).
We need to find public mobile or dense LiDAR or range data with imagery to go with it.
Possible sources:
What we need is a data folder with Makefile that pulls down the data and then processes into a form we can use.
I should be able to do this:
or just
The result should be that the datasets are downloaded (I suggest targets like
download-kitti) and then extracted (extract-kitti) and then split into batch-sized-chunks. Put the intermediate data in folders with names that uniquely depend on the way data is processed, likedata/intermediate/WxHxD-0xbfff/*.pkl, where0xbfffis supposed to be a hash of settings like the volume-min, volume-max, and any other config settings that might effect the output.Use Makefiles so that data only gets processed when it is absolutely needed.