Running separate models for each stream in parallel in DeepStream

Option 3 works