Issues loading data and loading pre-trained model about deep-flow-prediction HOT 2 CLOSED

Hi, L38 of dataset.py controls the mix of data from 3 directories. By default /reg, /shear, and /sup. E.g., you can pass [100, 0.7,0,0.3] which would mean there are 100 samples in total, and load 70 from "reg" and 30 from "sup".

The loading errors shouldn't happen. Maybe an older pytorch version could help here? We used version 4.1 originally. I hope that helps a bit!

from deep-flow-prediction.

I had the same missing key errors when loading the pre-trained models with load_dict_state. In DfpNet.py I replaced layer2b with layer2x. I guess the pre-trained models were trained with an older definition of TurbNetG.

from deep-flow-prediction.