Why the model size is much bigger than the origin video

[Monaco12138]

Thank you for your work. I tried to review your article and found that your model is much larger than the actual video. The model size is 30MB, but the actual video size is only less than 1MB. I am curious about any method to reduce the model size, as if the model is larger than the original video, this application scenario will be greatly limited

[lgupta-mle]

I would like to know the above as well. If the model's parameters are what are being transmitted, wouldn't such huge model parameter size cause issues? I simply checked the total size of the model parameters tensors and the NerV_S.pth checkpoint size added up to 12 MB. Compared to something like DCVC whose transmitting latent is around 127 KB, this seems like a big issue with NerV.

Please correct me if I am wrong!