Context must be initialized first

[fcogidi]

Hello,

I am working on a project to write videos to TFRecord format. I recently ran into an issue where I cannot get frames due to this error: Context must be initialized first. Here is a snippet of my code:

try:
  decord.bridge.set_bridge('tensorflow')
  vr = decord.VideoReader(path, ctx=decord.cpu(0))
  fps = math.ceil(vr.get_avg_fps())
  num_frames = min(len(vr), fps*10) # get at most 10 seconds of video
  frames = vr.get_batch(range(num_frames))
except Exception as e:
  logging.info(e)
  continue

This code is inside a loop, which is inside a function that is called by the multiprocessing.Pool.apply_async method. With a slight modification (removing the continue keyword), this code snippet works outside a loop, in a separate function. I need to be able to skip videos that can't be decoded, and it's easier to do that in a loop than in a function.

What could be going wrong and how do I get around it?

PS: I'm using TensorFlow==2.4.1 and decord==0.5.2

[zhreshold]

related to #132