# dataset_test = dset.ImageFolder(root=os.path.join(opt.dataset, opt.test_set), transform=transform_fwd)
# assert dataset_test
# dataloader_test = torch.utils.data.DataLoader(dataset_test, batch_size=opt.batchSize, shuffle=False, num_workers=int(opt.workers))
# dataloaders = {}
# for name in ['train', 'test']:
# raw_data = pandas.read_csv(os.path.join(opt.dataset, '%s.csv' % name))
# dataloaders[name] = DataLoader(FrameDataset(raw_data.to_numpy()), **config.dataset_params)
raw_data = pandas.read_csv(os.path.join(opt.dataset, 'test.csv'))
dataloader_test = DataLoader(FrameDataset(raw_data.to_numpy()),
batch_size=opt.batchSize,
shuffle=True,
num_workers=4,
pin_memory=False)
vgg_ext = model_big.VggExtractor()
capnet = model_big.CapsuleNet(2, opt.gpu_id)
capnet.load_state_dict(torch.load(os.path.join(opt.outf)))
capnet.eval()
if opt.gpu_id >= 0:
vgg_ext.cuda(opt.gpu_id)
capnet.cuda(opt.gpu_id)
##################################################################################
tol_label = np.array([], dtype=np.float)
tol_pred = np.array([], dtype=np.float)
tol_pred_prob = np.array([], dtype=np.float)
count = 0
print(path)
print('Number of files: %d' % (length))
print('Number of videos: %d' % (count_vid))
print('Number of correct classifications: %d' % (correct))
return count_vid, correct
if __name__ == '__main__':
path_real = os.path.join(opt.dataset, opt.real)
path_deepfakes = os.path.join(opt.dataset, opt.deepfakes)
path_face2face = os.path.join(opt.dataset, opt.face2face)
path_faceswap = os.path.join(opt.dataset, opt.faceswap)
vgg_ext = model_big.VggExtractor()
model = model_big.CapsuleNet(4, opt.gpu_id)
model.load_state_dict(
torch.load(os.path.join(opt.outf, 'capsule_' + str(opt.id) + '.pt')))
model.eval()
if opt.gpu_id >= 0:
vgg_ext.cuda(opt.gpu_id)
model.cuda(opt.gpu_id)
###################################################################
tol_count_vid_real = 0
tol_correct_real = 0
tol_count_vid_deepfakes = 0
tol_correct_deepfakes = 0