Sergio Arnaud

I had several issues running dopamine with [the last docker image](https://github.com/google/dopamine/blob/master/docker/core/Dockerfile) because of the last Jax and CUDA changes. Particularly with the cudnn versions. I made the following changes to...

Training is really slow

1

First of all, congratulations on the great work! I've been trying to train an agent to play breakout and the training is really slow. This is really confusing to me...

A couple of small bug fixes. - The first prevents an infinite loop while sampling receptacles - The second adds agent state to the debug visualizer which prevents an exception...

Indoor Scene Generation also generates outdoor terrain

2

The command provided in the readme to automatically generate indoor scenes takes too long (2 hours) the reason is that it's unecesarilly generating all the terrain of the outdoor scene,...

Could you advice me on which would be an ideal config file for my hardware? Bellow the specs: ``` GPUs: 8 GPU RAM: 32GB per GPU, 256GB in total GPU...

How to generate videos of indoor scenes?

3

I ran the following command ``` python infinigen.manage_jobs.py --output_folder outputs/go_big_go_home --num_scenes 32 --pipeline_configs local_256GB.gin monocular_video.gin indoor_background_configs.gin --configs singleroom.gin fast_solve.gin --pipeline_overrides get_cmd.driver_script=infinigen_examples.generate_indoors --overrides compose_indoors.restrict_single_supported_roomtype=True --overwrite ``` In particular notice that I'm...