Separate out densepose install for contributing instructions #179
Conversation
|
Okay @pjbull this is ready for review. Test failures are unrelated though I am looking into those |
|
Update: this PR now fixes the failing the test. The issue was that some of the test assets don't have that many frames since we heavily downsampled them -- if they have fewer than 10, then selecting the 10th frame index in the single frame test will fail. I added a sort in this commit to ensure we're always using the same video for that test which has enough frames to be a valid test case. |
Codecov Report
@@ Coverage Diff @@
## master #179 +/- ##
========================================
+ Coverage 85.0% 85.2% +0.1%
========================================
Files 30 30
Lines 1851 1851
========================================
+ Hits 1575 1578 +3
+ Misses 276 273 -3
|
ejm714
force-pushed
the
contributing-install
branch
from
7fe4d0b
to
137eeb4
Compare
| @@ -164,7 +164,7 @@ def test_train_save_dir_overwrite( | |||
| "train_configuration.yaml", | |||
| "test_metrics.json", | |||
| "val_metrics.json", | |||
| "epoch=0-step=10.ckpt", | |||
| "epoch=0-step=11.ckpt", | |||
There was a problem hiding this comment.
Caused by recent change in 1.6.0 (released 2022-03-29)
The filename of checkpoints created with ModelCheckpoint(filename='{step}') is different compared to previous versions. A checkpoint saved after 1 step will be named step=1.ckpt instead of step=0.ckpt (Lightning-AI/pytorch-lightning#11805)
requirements-dev.txtno longer installs the densepose extra since DensePose requires torch to be installed first. The appropriate install and test commands for DensePose contributions are now laid out on the contributing page.Bonus fixes
make testsruns with the default dev install. the verbose pytest will show that these tests are skipped