Replicating results from the original paper

I am trying to replicate results from the PhysDNet paper on the MovingMNIST dataset with the exact network parameters leading to a different MSE metric value. According to the paper [PhyDNet](https://arxiv.org/pdf/2003.01460.pdf), the MSE value should be around ~23 but after training using this package, I am getting MSE to be about ~380.
My question is 2 parts:
1. Have all the networks being tested to match the results for each of the models?
2. What could be the reason for the discrepancy in the results of PhyDNet?