I haved run this network and the accuracy I got is only 0.19. Is this normal?

Hi ,
I haved run this network and the accuracy I got is only 0.19. I want to ask is this result normal？
The log is below.
[epoch 97/100, episode 50/100] => loss: 2.65600, acc: 0.18667
[epoch 97/100, episode 100/100] => loss: 2.57098, acc: 0.21333
[epoch 98/100, episode 50/100] => loss: 2.66116, acc: 0.17000
[epoch 98/100, episode 100/100] => loss: 2.66873, acc: 0.18000
[epoch 99/100, episode 50/100] => loss: 2.58527, acc: 0.24667
[epoch 99/100, episode 100/100] => loss: 2.60647, acc: 0.20333
[epoch 100/100, episode 50/100] => loss: 2.50875, acc: 0.26000
[epoch 100/100, episode 100/100] => loss: 2.62107, acc: 0.18000
(20, 350, 84, 84, 3)
Testing...
[test episode 50/600] => loss: 1.73374, acc: 0.17333
[test episode 100/600] => loss: 1.73245, acc: 0.16000
[test episode 150/600] => loss: 1.65328, acc: 0.21333
[test episode 200/600] => loss: 1.68278, acc: 0.17333
[test episode 250/600] => loss: 1.64363, acc: 0.25333
[test episode 300/600] => loss: 1.77591, acc: 0.12000
[test episode 350/600] => loss: 1.58832, acc: 0.25333
[test episode 400/600] => loss: 1.66476, acc: 0.17333
[test episode 450/600] => loss: 1.71281, acc: 0.16000
[test episode 500/600] => loss: 1.64050, acc: 0.24000
[test episode 550/600] => loss: 1.71050, acc: 0.21333
[test episode 600/600] => loss: 1.72494, acc: 0.20000
Average Test Accuracy: 0.19098
-------------------------------------------------------
I didn't use download_miniimagenet.sh to download the dataset because of the 404 error. So I download the dataset ILSVRC2012_img_train_t3 from {http://www.image-net.org/challenges/LSVRC/2012/nonpub-downloads}. And I select 64 classes from the dataset and put them into the directory {prototypical-networks-tensorflow/data/mini-imagenet/data/train}, and 20 classes from the dataset and put them into the directory {prototypical-networks-tensorflow/data/mini-imagenet/data/test}. Does this create a problem?
PS: I run the .py file instead of .ipynb file.
Thank you so much! I am looking forward to your reply.