In [ ]:

    

%load_ext autoreload
%autoreload 2
import torch
from models.multimodal import load_data
import torch.utils.data as data
import torchvision.transforms as transforms
torch.multiprocessing.set_sharing_strategy('file_system')
import pickle


    

In [ ]:

    

reader_malo = load_data.MultimodalReader(
    '/disks/sdb1-3T/exportNFS/Databases/jobScreening_cvpr17/train/annotation_training.pkl', 
    '/disks/sdb1-3T/exportNFS/Databases/jobScreening_cvpr17/train/transcripts/ctms', 
    '/disks/sdb1-3T/exportNFS/Databases/jobScreening_cvpr17/train/audios/fbank', 
#     '/disks/sdb1-3T/exportNFS/csp/jobScreening_cvpr17/train/faces2/'
    '/disks/sdb1-3T/exportNFS/Databases/jobScreening_cvpr17/train/faces/vgg_features'
)


    

In [ ]:

    

normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
reader = load_data.MultimodalReader(
    '/disks/sdb1-3T/exportNFS/Databases/jobScreening_cvpr17/train/annotation_training.pkl', 
    '/disks/sdb1-3T/exportNFS/Databases/jobScreening_cvpr17/train/transcripts/ctms', 
    '/disks/sdb1-3T/exportNFS/Databases/jobScreening_cvpr17/train/audios/fbank', 
#     '/disks/sdb1-3T/exportNFS/csp/jobScreening_cvpr17/train/faces2/',
    '/disks/sdb1-3T/exportNFS/Databases/jobScreening_cvpr17/train/faces/vgg_features',
    transform=transforms.Compose([
                             transforms.Scale(240),
                             transforms.RandomSizedCrop(224),
                             transforms.RandomHorizontalFlip(),
                             transforms.ToTensor(),
                             normalize,
                         ]))


    

In [ ]:

    

train_loader_bad = data.DataLoader(
        reader, 
        batch_size=1, shuffle=True, num_workers=1)


    

In [ ]:

    

def my_collate2(batch):
    return batch


    

In [ ]:

    

print("Loading serialized_reader")
with open('serialized_reader.pickle','rb') as stream:
    reader = pickle.load(stream)    
print("Loaded serialized_reader")


    

In [ ]:

    

%reset_selective train_loader
train_loader = data.DataLoader(
    reader,
    batch_size=4,
    shuffle=True,
    num_workers=1,
    pin_memory=False, collate_fn=my_collate2)


    

In [ ]:

    

for mydata in train_loader:
    break


    

In [ ]:

    

print(len(mydata))

# with open('serialized_loaded_data.pickle','wb') as stream:
#     pickle.dump(mydata,stream)


    

In [ ]:

    

video=mydata[0][0][2]
video[0][0].shape[0]


    

In [ ]:

    

my_collate = load_data.my_collate
what = my_collate(mydata)


    

In [ ]:

    

mydata[3][0].size()


    

In [ ]:

    

load_data.my_collate


    

In [ ]:

    

from models.multimodal import seq_model


    

In [ ]: