# Copy ourselves to outdir to replicate results
if not os.path.exists(outdir):
os.makedirs(outdir)
shutil.copy(__file__, os.path.join(outdir, 'train.py'))
print('Loading data ... ', end='')
if rotate:
rotate_transform = datasets.RandomRotate(p=rotate)
else:
rotate_transform = None
dataset = {}
dataloader = {}
dataset['train'] = datasets.FreiburgRGBDDataset('data',
mode='train',
color=color,
transform=rotate_transform,
seed=42)
dataloader['train'] = DataLoader(dataset['train'],
batch_size=16,
shuffle=True,
num_workers=0)
dataset['val'] = datasets.FreiburgRGBDDataset('data',
mode='val',
color=color,
seed=42)
dataloader['val'] = DataLoader(dataset['val'],
batch_size=16,
shuffle=True,
num_workers=0)
print('OK.')
# Copy ourselves to outdir to replicate results
if not os.path.exists(outdir):
os.makedirs(outdir)
shutil.copy(__file__, os.path.join(outdir, 'train.py'))
print('Loading data ... ', end='')
if rotate:
rotate_transform = datasets.RandomRotate(p=rotate)
else:
rotate_transform = None
dataset = {}
dataloader = {}
dataset['train'] = datasets.FreiburgRGBDDataset('data',
mode='train',
color=None,
transform=None,
seed=42)
dataloader['train'] = DataLoader(dataset['train'],
batch_size=16,
shuffle=True,
num_workers=0)
dataset['val'] = datasets.FreiburgRGBDDataset('data',
mode='val',
color=None,
seed=42)
dataloader['val'] = DataLoader(dataset['val'],
batch_size=16,
shuffle=True,
num_workers=0)
print('OK.')
import numpy as np
import shutil
from torch.autograd import Variable
import torch.nn as nn
import torch.optim as optim
import datasets
import networks
# Variables
dropout = True
color = True
outdir = '/imatge/jdelarica/work/PycharmProjects/TFG/tests_512_100_50_1/sgd_lr0.0001_mom0.0_wd0.0_drop_color'
print('Loading data ... ', end='')
dataset = datasets.FreiburgRGBDDataset('data', mode='test', color=color)
dataloader = DataLoader(dataset, batch_size=16, shuffle=True, num_workers=0)
print('OK.')
print('Creating network ... ', end='')
net = networks.NetFC(nodes=[dataset.get_input_size(), 600, 100, 1],
dropout=dropout)
net.load_state_dict(torch.load(os.path.join(outdir, 'net.pkl')))
print('OK.')
print('Creating loss and optimizer ... ', end='')
criterion = nn.BCELoss(size_average=False)
#optimizer = optim.SGD(net.parameters(), lr=lr, momentum=momentum, weight_decay=weight_decay)
print('OK.')
cuda = torch.cuda.is_available()
