This PR should be squashed, a lot of commits are non significant (CI triggers).
The implemented workflow is the following:
On new PR opened
1) Train new model on all datasets
2) Compute model statistics on ground truth repository
3) Compare statistics and show results in PR comment
On PR merged
1) Version bump
2) New release that contains the new trained model
In order to work, it needs the following setup:
GROUND_TRUTH_REPO_TOKEN: secret containing github token with permissions to read/write public and private repos
GROUND_TRUTH_REPO: secret containing the url of the private ground truth repo
GITHUB_TOKEN: default token needs Read and Write permissions (repository or organization Settings -> Actions -> Read and Write)
File structure: one folder one model (las, laz or ply) + optional json descriptor
Every model in the ground truth repository should be paired with a stats.json file (generated with pcclassify)
This PR should be squashed, a lot of commits are non significant (CI triggers).
The implemented workflow is the following:
On new PR opened 1) Train new model on all datasets 2) Compute model statistics on ground truth repository 3) Compare statistics and show results in PR comment
On PR merged 1) Version bump 2) New release that contains the new trained model
In order to work, it needs the following setup:
GROUND_TRUTH_REPO_TOKEN
: secret containing github token with permissions to read/write public and private reposGROUND_TRUTH_REPO
: secret containing the url of the private ground truth repoGITHUB_TOKEN
: default token needsRead and Write
permissions (repository or organizationSettings -> Actions -> Read and Write
)las
,laz
orply
) + optional json descriptorstats.json
file (generated withpcclassify
)