Kaggle + VecVideoRecorder + VecFrameStack

I am trying to record a video of a stacked VecEnv.

test_env = make_atari_env(atari_env_name, n_envs=1, seed=0)
test_env = VecFrameStack(test_env, n_stack=4)

This is the record_video function:

from stable_baselines3.common.vec_env import VecVideoRecorder, DummyVecEnv

def record_video(eval_env, model, video_length=500, prefix='', video_folder=video_folder):
  """
  :param env_id: (str)
  :param model: (RL model)
  :param video_length: (int)
  :param prefix: (str)
  :param video_folder: (str)
  """
  # Start the video at step=0 and record 500 steps
  eval_env = VecVideoRecorder(eval_env, video_folder=video_folder,
                              record_video_trigger=lambda step: step == 0, video_length=video_length,
                              name_prefix=prefix)

  obs = eval_env.reset()
  for _ in range(video_length):
    action, _ = model.predict(obs)
    obs, _, _, _ = eval_env.step(action)

  # Close the video recorder
  eval_env.close()

and I run

record_video(test_env, model, video_length=10000, prefix='a2c_pong')

VecVideoRecorder generates the mp4 file

Saving video to /kaggle/working/videos/a2c_pong-step-0-to-step-10000.mp4

but the file is always just 262 bytes long, regardles of video_length

total 16
drwxr-xr-x 2 root root 4096 Feb 12 16:37 .
drwxr-xr-x 5 root root 4096 Feb 12 16:37 ..
-rw-r--r-- 1 root root 1448 Feb 12 16:43 a2c_pong-step-0-to-step-10000.meta.json
-rw-r--r-- 1 root root  262 Feb 12 16:43 a2c_pong-step-0-to-step-10000.mp4

and the mp4 wont play. Is this an issue with the stacked VecEnv?

[araffin]

Hello,

Please fill in the issue template completely (notably your gym version here, may be related to https://github.com/DLR-RM/stable-baselines3/issues/294).

Have you tried with the script from the rl zoo? https://github.com/DLR-RM/rl-baselines3-zoo

[araffin]

solution as for now is to downgrade gym to 0.17.3

or to use gym master version.