the best valid acc is 87% which is close to 90%
first download the dataset and unzip it to test*
dir and then modify the data_path
item and then
python train.py
Note!
I'm sorry, I have little understand about how to use the physical equation to the picture data
But when I reproducing the Lensiformer model I find that the
loss
went up rather than down
, and I did not figure out where does he apply the physical formula