Pytorch for handwritten grapheme classification

dataset banglali.ai from: https://www.kaggle.com/c/bengaliai-cv19

This problem is multi lable classification.

models.py:

pretrain models: https://github.com/Cadene/pretrained-models.pytorch

use resnet34: https://github.com/Cadene/pretrained-models.pytorch/blob/master/pretrainedmodels/models/torchvision_models.py
--->>> line 358: def resnet34(num_classes=1000, pretrained='imagenet'): --->>> line 314: def modify_resnets(model):

earlystopping: https://github.com/Bjarten/early-stopping-pytorch/blob/master/pytorchtools.py

Inference