Pablo Villanueva Domingo
Pablo Villanueva Domingo
Great, thanks for the response!
Hello @DaniCarias , Are you using the same values for `img_width` and `img_height`? In your last response you made use of `focal_length` for `fx` and `fy`, which should be different...
I'm not sure what the open3d routine is doing under the hood, but (as is also stated [here](https://github.com/carla-simulator/carla/issues/56#issuecomment-347957348)) you could convert a 2D point `p2d = [u,v,1]` to the 3d...
Hi, thanks for your question. Scenario runner only provides routes for Town12 as training routes (`routes_training.xml`), and for Town13 as validation routes (`routes_validation.xml`). Note that there is also a `routes_devtest.xml`...
Thanks for the PR. Given that it is a static class initialized at the beginning, we can ignore `None`s. Therefore, we prefer the Practical B choice.
We agree on that change and proceed to merge it. Thanks!
But is this behavior expected in the SAC implementation of sb3? What do you mean with "save/load the replay buffer too"? Thanks!
@qgallouedec thanks for your answer! But even saving and loading the buffer, when resuming training the mean reward starts from the low initial value. Is that expected? The orange line...
Thanks for the PR. Could you please do the PR to the master branch and change the minimum version to 0.9.15, which is the default version used in Leaderboard 2?
@NinePeAkSword To load a model previously trained, you can just load it with `trainer.load("10")`, which loads the 10-th checkpoint in the results folder.