In case of visual observation inputs, how can we add images to tensorboard summary so we can understand what the model is seeing?
Thanks.
Tensoboard logs should include your observations as images if you initialize your agent with full_tensorboard_log=True (see e.g. PPO2 docs). Note that this will quickly eat up space, so you might not want to train anything for a long while with this enabled.
In the code, it is done here: https://github.com/hill-a/stable-baselines/blob/cf2e19cb42e51361d30ddcdbad5b88e607567ca6/stable_baselines/ppo2/ppo2.py#L232-L233
Thanks, that's it.
Most helpful comment
In the code, it is done here: https://github.com/hill-a/stable-baselines/blob/cf2e19cb42e51361d30ddcdbad5b88e607567ca6/stable_baselines/ppo2/ppo2.py#L232-L233