try_resnext101_32x4d.py

from models.resnext import resnext101_32x4d
from iceberg import *
from torch.optim import SGD

lr = 0.0005
mom = 0.9
wd = 1e-4

model = resnext101_32x4d()
model = torch.nn.DataParallel(model).cuda()
optim = SGD(model.parameters(), lr, momentum=mom, weight_decay=wd)
iceberg = Iceberg('./data/train.json', './data/resnext101_32x4d_lr5e-4_mom9e-1_wd1e-4_bs256_pretrained_model.pth', model, optim, 100000, 128)
iceberg.run(transforms.Compose([
            transforms.Resize(256),
            transforms.RandomCrop(224),
            transforms.RandomHorizontalFlip(),
            transforms.ToTensor()]))