x = self.relu(x)
x = self.maxpool(x)
x = self.layer1(x)
x = self.layer2(x)
x = self.layer3(x)
x = self.layer4(x)
x = self.avgpool(x)
x = x.view(x.size(0), -1)
# x = self.fc(x)
return x
data_set = INaturalistDataset(
base_folder + "data_preprocessed_" + str(size) + "/",
base_folder + "annotations/train2017_vis.json", transform, False)
data_loader = torch.utils.data.DataLoader(data_set,
batch_size=batch_size,
shuffle=True)
model = torch.load(open(model_file, "rb"))
model.cuda = False
model.forward = new_forward
N = len(data_loader) * batch_size
X = np.zeros((N, 2048))
y = np.zeros((N, ))
for i, (data, targets) in enumerate(data_loader):
from preprocessing.inaturalist_dataset import INaturalistDataset
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torchvision import transforms
from torch.autograd import Variable
data_dir = './data_min2/'
annotations_dir = './annotations/modular_network/Mammalia/'
train_annotations = '{}train2017_min.json'.format(annotations_dir)
val_annotations = '{}val2017_min.json'.format(annotations_dir)
inaturalist = INaturalistDataset(data_dir,
train_annotations,
transform=transforms.ToTensor())
all_ids = inaturalist.all_ids
# images, targets = inaturalist.get_images(all_ids)
batch_size = 10
train_loader = torch.utils.data.DataLoader(inaturalist,
batch_size=batch_size,
shuffle=True)
class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
self.conv1 = nn.Conv2d(3, 5, kernel_size=10)
self.conv2 = nn.Conv2d(5, 10, kernel_size=10)
epochs = 1
log_interval = 10
loss = nn.CrossEntropyLoss()
output_categories = 3
optimizer = optim.Adam
# set directories
data_dir = './data_min_preprocessed_224/'
annotations_dir = './annotations/modular_network/Mammalia/'
train_annotations = '{}train2017_min.json'.format(annotations_dir)
val_annotations = '{}val2017_min.json'.format(annotations_dir)
# create data sets
applied_transformation = transforms.Compose([transforms.ToTensor()])
inaturalist_train = INaturalistDataset(data_dir,
train_annotations,
transform=applied_transformation,
modular_network_remap=False)
inaturalist_val = INaturalistDataset(data_dir,
val_annotations,
transform=applied_transformation,
modular_network_remap=False)
# create loaders for the data sets
train_loader = torch.utils.data.DataLoader(inaturalist_train,
batch_size=batch_size,
shuffle=True)
val_loader = torch.utils.data.DataLoader(inaturalist_val,
batch_size=batch_size,
shuffle=True)
# get pre-trained model, change FC layer
100. * batch_idx * len(data)/ len(dataset_loader.dataset)),
end='\r')
loss_value /= len(val_loader.dataset)
# final log
print('\nEvaluation results:\nAverage loss: {:.4f}, Accuracy: {}/{} ({:.2f}%)\n'.format(
loss_value, correct, len(dataset_loader.dataset),
100. * correct / len(dataset_loader.dataset)))
if __name__ == '__main__':
# training
print("\n\nLoading training set...")
inaturalist_train = INaturalistDataset(data_dir, train_annotations, transform=None,
modular_network_remap=False)
labels = numpy.zeros(len(inaturalist_train))
stacked = numpy.zeros((len(inaturalist_train),268203))
for i, photo_data in enumerate(inaturalist_train):
#print(numpy.shape(numpy.array(photo_data[0]).flatten()))
stacked[i] = numpy.array(photo_data[0]).flatten()
labels[i] = photo_data[1][1]
#for component in (photo_data[1]):
# print("Class: ", component)
print(numpy.shape(stacked))
print(stacked[0])
print("Labels: ", labels)
"""
train_loader = torch.utils.data.DataLoader(inaturalist_train, batch_size=batch_size, shuffle=True)
print("Starting training (%d epochs)" % epochs)
for epoch_count in range(1, epochs + 1):
model_parameters[parameter_name] = parameter_value
return torch.load(open(models_base_folder + filename, "rb")), model_parameters
if __name__ == '__main__':
print("Loading models...")
models = []
for path in pth_filenames:
model, parameters = load_model(path)
models.append(model)
print("done.")
# print(type(model))
# produce test loader
input_size = 224
inaturalist_test = INaturalistDataset(test_dir(input_size), test_annotations,
transform=applied_transformations,
modular_network_remap=False)
test_loader = torch.utils.data.DataLoader(inaturalist_test, batch_size=test_batch_size)
# find loss
if parameters['loss'] == 'CrossEntropyLoss':
loss = torch.nn.CrossEntropyLoss()
else:
raise ValueError("Invalid loss '%s'" % parameters['loss'])
evaluate(models, loss, test_loader)