About the ablation study

[Tung-I]

Hi,

I am wondering how to reporduce the reported ablation study results in the paper. Specifically, in Tab. 3, the performance of "PARIMA", "PA", and "ARIMA" are provided, where (based on my understanding) Parima is the proposed approach leverages both viewport and object trajectories, PA/ARIMA are the baselines using object trajectories/viewport trajectories only.

Could you please kindly provide the sample implementation of the two baselines, PA and ARIMA as well if possible? It will be greatly helpful to evaluate and quantify the impact brought by the video semantics. Thanks.