Erik Wijmans
Erik Wijmans
I'm not familiar with Open3D, but `depth_scale=1000.0` seems odd. Habitat-lab outputs depth in `[0,1]` by default where a value of 1 represents something 10m away, so maybe depth_scale=10.0? 2. Not...
We current support the Redwood depth noise model (http://redwood-data.org/indoor/dataset.html). That can be added by `config.SIMULATOR.DEPTH_SENSOR.NOISE_MODEL = "RedwoodDepthNoiseModel"`. The easiest way to achieve the FOV you want is to set the...
> newer checkpoint files contain a config entry, older checkpoint files do not. The configs have been removed from released models to not tie those weights to a specific version...
> As for the memory issue - this might happen with a big dataset such as matterport or gibson i guess ? The habitat test scenes dataset should not be...
I am not sure what is going on. Does the RL training part of this colab work: https://colab.research.google.com/github/facebookresearch/habitat-lab/blob/master/examples/tutorials/colabs/Habitat_Lab.ipynb ? Also make sure you are requesting a GPU instance.
Hmm, odd. Maybe there is something with !python in colab then. `run.py` is pretty simple and you even grab `run_exp` from it, so maybe things will work if you just...
It looks like you are calling into habitat-baselines via ``` !python -u habitat_baselines/run.py --exp-config habitat_baselines/config/pointnav/ppo_pointnav_example.yaml --run-type train ``` The example colab doesn't do that, it just invokes the trainers directly...
The normal there unfortunately won't be useful when the agent is in contact with a surface as it is just `normalized(agent_position - hit_position)`. There also isn't an easy way to...
This would require a new feature.
``` agent_state.position = np.array([-0.6, 0.0, 0.0]) ``` This demo is placing the agent in a fixed location. This location won't be appropriate for all scenes. You can use `agent_state.position =...