def main():
# Load config
config = get_config_from_json('config/model.config')
# Load data
[X_train, Y_train, X_CV, Y_CV, X_test, Y_test] = load_data(0.18)
# Generate dataset
test_dataset = MyDataset(X_test, Y_test)
# Create Data Loaders
test_loader = DataLoader(dataset=test_dataset,
batch_size=config.model.batch_size,
shuffle=False)
# Build model
model = CNNModel()
model = model.double()
model.eval()
if os.path.isfile(config.model.path):
model.load_state_dict(torch.load(config.model.path))
print('Loaded checkpoint..')
else:
print('checkpoint not found..')
evaluate(model, test_loader)
def on_epoch_end(self, epoch, logs=None):
average_precisions = evaluate(
self.model_to_save,
self.valid_generator,
net_h=self.config["model"]["min_input_size"],
net_w=self.config["model"]["min_input_size"])
log_message = ""
log_message += "epoch : " + str(epoch + 1) + "\n"
# print the score
for label, average_precision in average_precisions.items():
log_message += self.config["model"]["labels"][
label] + ': {:.4f}'.format(average_precision) + "\n"
print(self.config["model"]["labels"][label] +
': {:.4f}'.format(average_precision))
print('mAP: {:.4f}'.format(
sum(average_precisions.values()) / len(average_precisions)))
log_message += 'mAP: {:.4f}'.format(
sum(average_precisions.values()) /
len(average_precisions)) + "\n\n"
log_message += "==============================================\n\n"
f = open(os.path.join("./", self.train_mAP, self.log_file), "a+")
f.write(log_message)
f.close()
super(My_Checkpoint, self).on_batch_end(epoch, logs)
def fit_model():
n_clusters = K
print('')
print('=' * 50)
print("Let's start with {} clusters!".format(n_clusters))
print('=' * 50)
print('')
train_test_clu, train_test_reg, df_weather = read_dataset()
clusterer = create_clusterer(n_clusters)
df_clustered_list, sequence, label_df_test = clusterer.fit(train_test_clu)
classifier = create_classifier()
best_classifier, test_clf = classifier.fit(train_test_clu, sequence,
df_weather)
regressor = create_regressor()
best_model_set = regressor.fit(df_clustered_list, train_test_reg)
pred_for_plot, true_for_plot = evaluate(label_df_test, train_test_reg,
test_clf, best_classifier,
best_model_set, sequence)
return
def on_epoch_end(self, epoch, logs=None):
average_precisions = evaluate(self.infer_model, self.valid_generator)
print(f'Epoch {epoch} Validation Metrics:')
print(' mAP: {:.4f}'.format(
sum(average_precisions.values()) / len(average_precisions)))
for label, average_precision in average_precisions.items():
print(' ' + self.labels[label] +
': {:.4f}'.format(average_precision))
def _main_(args):
config_path = args.conf
with open(config_path) as config_buffer:
config = json.loads(config_buffer.read())
print(config['valid']['valid_annot_folder'])
###############################
# Create the validation generator
###############################
valid_ints, labels = parse_voc_annotation(
config['valid']['valid_annot_folder'],
config['valid']['valid_image_folder'], config['valid']['cache_name'],
config['model']['labels'])
labels = labels.keys() if len(
config['model']['labels']) == 0 else config['model']['labels']
labels = sorted(labels)
valid_generator = BatchGenerator(
instances=valid_ints,
anchors=config['model']['anchors'],
labels=labels,
downsample=
32, # ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image=0,
batch_size=config['train']['batch_size'],
min_net_size=config['model']['min_input_size'],
max_net_size=config['model']['max_input_size'],
shuffle=True,
jitter=0.0,
norm=normalize)
###############################
# Load the model and do evaluation
###############################
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
infer_model = load_model(config['train']['saved_weights_name'])
# compute mAP for all the classes
average_precisions = evaluate(infer_model,
valid_generator,
iou_threshold=iou_threshold,
obj_thresh=obj_thresh,
nms_thresh=nms_thresh,
net_h=net_h,
net_w=net_w)
# print("recall : {}".format(recall))
# print("precision : {}".format(precision))
# plt.plot(recall, precision)
# plt.show()
# print the score
for label, average_precision in average_precisions.items():
print(labels[label] + ': {:.4f}'.format(average_precision))
print('mAP: {:.4f}'.format(
sum(average_precisions.values()) / len(average_precisions)))
def train_rnn_model(train_loader, val_loader, num_features, num_epochs,
use_cuda, path_output):
"""
Use train and validation loader to train the variable RNN model
Input: train_loader, val_loader
Output: trained best model
"""
device = torch.device(
"cuda" if torch.cuda.is_available() and use_cuda else "cpu")
torch.manual_seed(1)
if device.type == "cuda":
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
model = VariableRNN(num_features)
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters())
model.to(device)
criterion.to(device)
best_val_acc = 0.0
train_losses, train_accuracies = [], []
valid_losses, valid_accuracies = [], []
for epoch in range(num_epochs):
train_loss, train_accuracy = train(model, device, train_loader,
criterion, optimizer, epoch)
valid_loss, valid_accuracy, valid_results = evaluate(
model, device, val_loader, criterion)
train_losses.append(train_loss)
valid_losses.append(valid_loss)
train_accuracies.append(train_accuracy)
valid_accuracies.append(valid_accuracy)
is_best = valid_accuracy > best_val_acc
if is_best:
best_val_acc = valid_accuracy
torch.save(
model,
os.path.join(path_output, "VariableRNN.pth"),
_use_new_zipfile_serialization=False,
)
best_model = torch.load(os.path.join(path_output, "VariableRNN.pth"))
return (
best_model,
train_losses,
valid_losses,
train_accuracies,
valid_accuracies,
valid_results,
)
def _main_(args):
config_path = args.conf
with open(config_path) as config_buffer:
config = json.loads(config_buffer.read())
###############################
# Create the validation generator
###############################
valid_ints, labels = parse_voc_annotation(
config['valid']['valid_annot_folder'],
config['valid']['valid_image_folder'], config['valid']['cache_name'],
config['model']['labels'])
labels = labels.keys() if len(
config['model']['labels']) == 0 else config['model']['labels']
labels = sorted(labels)
valid_generator = BatchGenerator(
instances=valid_ints,
anchors=config['model']['anchors'],
labels=labels,
downsample=
32, # ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image=0,
batch_size=config['train']['batch_size'],
min_net_size=config['model']['min_input_size'],
max_net_size=config['model']['max_input_size'],
shuffle=True,
jitter=0.0,
norm=normalize)
###############################
# Load the model and do evaluation
###############################
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
h5_files = []
for root, dirs, files in os.walk('.'):
for weights in files:
if weights[-3:] == '.h5':
if weights[:6] == 'helmet':
h5_files.append(weights)
print(h5_files)
for i in h5_files:
infer_model = load_model(i)
# compute mAP for all the classes
average_precisions = evaluate(infer_model, valid_generator)
# print the score
for label, average_precision in average_precisions.items():
print(labels[label] + ': {:.4f}'.format(average_precision))
print('mAP: {:.4f} of weight {}'.format(
sum(average_precisions.values()) / len(average_precisions), i))
def evaluate_main_(args):
config_path = args.conf
yolo_config_file_exit('pass') \
if os.path.isfile(config_path) \
else yolo_config_file_exit('fail')
with open(config_path) as config_buffer:
config = json.loads(config_buffer.read())
###############################
# Create the validation generator
###############################
valid_ints, labels = parse_voc_annotation(
config['valid']['valid_annot_folder'],
config['valid']['valid_image_folder'], config['valid']['cache_name'],
config['model']['labels'])
labels = labels.keys() if len(config['model']['labels']) == 0 \
else config['model']['labels']
labels = sorted(labels)
valid_generator = BatchGenerator(
instances=valid_ints,
anchors=config['model']['anchors'],
labels=labels,
downsample=32,
# ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image=0,
batch_size=config['train']['batch_size'],
min_net_size=config['model']['min_input_size'],
max_net_size=config['model']['max_input_size'],
shuffle=True,
jitter=0.0,
norm=normalize)
###############################
# Load the model and do evaluation
###############################
#os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
infer_model = load_model(config['train']['saved_weights_name'])
# compute mAP for all the classes
average_precisions = evaluate(infer_model, valid_generator)
# print the score
for label, average_precision in average_precisions.items():
print(labels[label] + ': {:.4f}'.format(average_precision))
print('mAP: {:.4f}'.format(
sum(average_precisions.values()) / len(average_precisions)))
try:
yolo_process_exit('pass')
except RuntimeError as e:
yolo_process_exit('fail')
def validate(self):
"""
One cycle of model validation
:return:
"""
self.testmodel.load_state_dict(self.model.state_dict())
self.testmodel.eval()
gpu_device = self.config.gpu_device
if self.config.data_loader == 'Isc':
nmi, train_recall = evaluate(gpu_device,
self.testmodel,
self.data_loader.train_loader,
self.config.train_classes,
name='tra_similar.jpg')
nmi, recall = QG_evaluate(gpu_device,
self.testmodel,
self.data_loader.query_loader,
self.data_loader.gallery_loader,
self.config.test_classes,
name='tes_similar.jpg')
else:
nmi, train_recall = evaluate(gpu_device,
self.testmodel,
self.data_loader.valid_loader,
self.config.train_classes,
name='tra_similar.jpg')
nmi, recall = evaluate(gpu_device,
self.testmodel,
self.data_loader.test_loader,
self.config.test_classes,
name='tes_similar.jpg')
self.logger.info("**Evaluating...**")
self.logger.info("NMI: {:.3f}".format(nmi * 100))
if nmi != 0:
for i, k in enumerate([1, 2, 4, 8]):
self.logger.info("R@{} : {:.3f}".format(k, 100 * recall[i]))
return train_recall[0], recall[0]
else:
for i, k in enumerate([1, 10, 20, 30]):
self.logger.info("R@{} : {:.3f}".format(k, 100 * recall[i]))
return train_recall[0], recall[0]
def _main_(args):
config_path = args.conf
with open(config_path) as config_buffer:
config = json.loads(config_buffer.read())
###############################
# Create the validation generator
###############################
valid_ints, labels = parse_voc_annotation(
config['valid']['valid_annot_folder'],
config['valid']['valid_image_folder'], config['valid']['cache_name'],
config['model']['labels'])
labels = labels.keys() if len(
config['model']['labels']) == 0 else config['model']['labels']
labels = sorted(labels)
valid_generator = BatchGenerator(
instances=valid_ints,
anchors=config['model']['anchors'],
labels=labels,
downsample=
32, # ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image=0,
batch_size=config['train']['batch_size'],
min_net_size=config['model']['min_input_size'],
max_net_size=config['model']['max_input_size'],
shuffle=True,
jitter=0.0,
norm=normalize)
###############################
# Load the model and do evaluation
###############################
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
infer_model = load_model(config['train']['saved_weights_name'])
# compute mAP for all the classes
average_precisions, recall, precision = evaluate(infer_model,
valid_generator)
# print the score
for (c, ap), (c, prec), (c, call) in zip(average_precisions.items(),
precision.items(),
recall.items()):
print(
"+ Class {c} - AP: {ap}, precision: {prec}, recall: {call}".format(
c=c, ap=ap, prec=prec, call=call))
map = np.mean(list(average_precisions.values()))
mprec = np.mean(list(precision.values()))
mrecall = np.mean(list(recall.values()))
print("mAP: {map}, mprec: {mprec}, mrecall: {mrecall}".format(
map=map, mprec=mprec, mrecall=mrecall))
def main(argv):
C0 = 0
C1 = 8
# Read args from command line
sampleSize = utils.parseArgs(argv)
# Load the train and test sets from MNIST
print("Loading datasets from MNIST...")
(x_train, y_train), (x_test, y_test) = mnist.load_data()
# Apply preprocessing to the training and test sets
print("Preprocessing training set...")
x_train, y_train = utils.preprocess(x_train, y_train, C0, C1)
print("Preprocessing testing set...")
x_test, y_test = utils.preprocess(x_test, y_test, C0, C1)
# Apply feature selection to training set
print("Applying feature selection...")
x_train, x_test = utils.featureSelection(x_train, x_test)
# Split training set by class
x0_train = [_ for i, _ in enumerate(x_train) if y_train[i] == 0]
x1_train = [_ for i, _ in enumerate(x_train) if y_train[i] == 1]
# Take random sample of each class of training set
print("Sampling {}% of training set".format(sampleSize * 100))
x0_train_sample = random.sample(x0_train, int(len(x0_train) * sampleSize))
x1_train_sample = random.sample(x1_train, int(len(x1_train) * sampleSize))
# Use Dr Arodz's code to get MAP estimate
print(
"Running Dr Arodz's code to obtain MAP estimates of means and covariance"
)
m0, m1, cov = Arodz(x0_train_sample, x1_train_sample)
# Predict labels for test set
print("Testing model...")
labels = predict(m0, m1, cov, x_test)
# Evaluate label accuracy
utils.evaluate(labels, y_test)
def main(_):
image_path = FLAGS.test
csv_path = os.path.splitext(image_path)[0] + ".csv"
# --------- load classifier ------- #
cascade = cv2.CascadeClassifier(FLAGS.cascade_xml)
model, x, keep_prob = get_nn_classifier()
# ---------- object detection ------------#
print 'starting detection of ' + FLAGS.test + '...'
img = utils.getImage(image_path)
img = cv2.normalize(img, None, 0, 255, cv2.NORM_MINMAX, cv2.CV_8U)
delta = [-2, -1, 0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.85, 0.9, 0.95, 0.99, 0.995, 0.999, 0.9995, 0.9999]
start = time.time()
candidates = cascade.detectMultiScale(img, scaleFactor=FLAGS.scaleFactor, minNeighbors=FLAGS.minNeighbors, maxSize=(FLAGS.max_window_size,FLAGS.max_window_size))
detected = nn_classification(candidates, img, model, x, keep_prob, delta)
elapsed = (time.time() - start)
print 'detection time: %d' % elapsed
# ------------- evaluation --------------#
ground_truth_data = utils.get_ground_truth_data(csv_path)
for j in xrange(0, len(delta)):
detected[j] = [Rect(x, y, w, h) for (x,y,w,h) in detected[j]]
tp, fn, fp = utils.evaluate(ground_truth_data, detected[j])
# ----------------output ----------------#
# image output
"""
img_out = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB)
for (x,y,w,h) in detected[j]:
cv2.rectangle(img_out, (x-w/2,y-h/2),(x+w/2,y+h/2), [0,255,0], 3)
for c in ground_truth_data:
cv2.circle(img_out, (c[0], c[1]), 3, [0,0,255],3)
output_file = "out" + '_' + str(datetime.datetime.now())
cv2.imwrite(FLAGS.output_dir + output_file + '.png', img_out)
"""
# csv output
with open(FLAGS.output_dir + FLAGS.out + '.csv', 'ab') as file:
writer = csv.writer(file, delimiter=',')
writer.writerow([FLAGS.test, str(elapsed), str(len(ground_truth_data)), delta[j], FLAGS.minNeighbors, FLAGS.scaleFactor,
str(len(detected[j])), str(tp), str(fp), str(fn)])
def main(argv):
C0 = 0
C1 = 8
# Read args from command line
sampleSize = utils.parseArgs(argv)
# Load the train and test sets from MNIST
print("Loading datasets from MNIST...")
(x_train, y_train), (x_test, y_test) = mnist.load_data()
# Apply preprocessing to the training and test sets
print("Preprocessing training set...")
x_train, y_train = utils.preprocess(x_train, y_train, C0, C1)
print("Preprocessing testing set...")
x_test, y_test = utils.preprocess(x_test, y_test, C0, C1)
# Apply feature selection to training set
# print("Applying feature selection...")
# x_train, x_test = utils.featureSelection(x_train, x_test)
# Sample training set
sampleIndicies = random.sample(range(len(x_train)), int(len(x_train)*sampleSize))
x_train_sample = [_ for i, _ in enumerate(x_train) if i in sampleIndicies]
y_train_sample = [_ for i, _ in enumerate(y_train) if i in sampleIndicies]
# Obtain MAP estimates
print("Running Dr Arodz's code to obtain MAP estimates of w and b")
w, b = Arodz(x_train_sample, y_train_sample)
# Predict labels for test set
print("Testing model...")
labels = predict(w, b, x_test)
# Evaluate label accuracy
utils.evaluate(labels, y_test)
def _main_(args):
config_path = args.conf
with open(config_path) as config_buffer:
config = json.loads(config_buffer.read())
###############################
# Create the validation generator
###############################
valid_ints, labels = parse_voc_annotation(
config['valid']['valid_annot_folder'],
config['valid']['valid_image_folder'],
config['valid']['cache_name'],
config['model']['labels']
)
labels = labels.keys() if len(config['model']['labels']) == 0 else config['model']['labels']
labels = sorted(labels)
valid_generator = BatchGenerator(
instances = valid_ints,
anchors = config['model']['anchors'],
labels = labels,
downsample = 32, # ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image = 0,
batch_size = config['train']['batch_size'],
min_net_size = config['model']['min_input_size'],
max_net_size = config['model']['max_input_size'],
shuffle = True,
jitter = 0.0,
norm = normalize
)
###############################
# Load the model and do evaluation
###############################
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
infer_model = load_model(config['train']['saved_weights_name'])
# compute mAP for all the classes
average_precisions = evaluate(infer_model, valid_generator)
# print the score
for label, average_precision in average_precisions.items():
print(labels[label] + ': {:.4f}'.format(average_precision))
print('mAP: {:.4f}'.format(sum(average_precisions.values()) / len(average_precisions)))
def main():
image_path = FLAGS.test
csv_path = os.path.splitext(image_path)[0] + ".csv"
# ------------ load classifier ---------- #
cascade = cv2.CascadeClassifier(FLAGS.cascade_xml)
# -------------- open image --------------#
img = utils.getImage(image_path)
img = cv2.normalize(img, None, 0, 255, cv2.NORM_MINMAX, cv2.CV_8U)
# ---------- object detection ------------#
print 'starting detection of ' + FLAGS.test + '...'
start = time.time()
detected = cascade.detectMultiScale(img, scaleFactor=FLAGS.scaleFactor, minNeighbors=FLAGS.minNeighbors, maxSize=(FLAGS.max_window_size, FLAGS.max_window_size))
elapsed = (time.time() - start)
print 'detection time: %d' % elapsed
# ------------- evaluation --------------#
detected = [Rect(x, y, w, h) for (x,y,w,h) in detected]
ground_truth_data = utils.get_ground_truth_data(csv_path)
tp, fn, fp = utils.evaluate(ground_truth_data, detected)
# ----------------output ----------------#
# image output
"""
img_out = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB)
for c in ground_truth_data:
cv2.circle(img_out, (c[0], c[1]), 3, [0,0,255],3)
for r in detected:
cv2.rectangle(img_out, (r.x, r.y), (r.x2(), r.y2()), [0,255,0], 2)
output_file = "out" + '_' + str(datetime.datetime.now())
cv2.imwrite(FLAGS.output_dir + output_file + '.png', img_out)
"""
# csv output
with open(FLAGS.output_dir + 'results.csv', 'ab') as file:
writer = csv.writer(file, delimiter=',')
writer.writerow([FLAGS.test, str(elapsed),str(len(ground_truth_data)), str(FLAGS.scaleFactor),
str(FLAGS.minNeighbors), str(len(detected)), str(tp), str(fp), str(fn)])
def _main_(args):
config_path = args.conf
with open(config_path) as config_buffer:
config = json.loads(config_buffer.read())
###############################
# Create the validation generator
###############################
valid_ints, labels = parse_voc_annotation(
config['valid']['valid_annot_folder'],
config['valid']['valid_image_folder'], config['model']['labels'])
labels = sorted(labels.keys())
valid_generator = BatchGenerator(
instances=valid_ints,
anchors=config['model']['anchors'],
labels=labels,
downsample=
32, # ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image=config['model']['max_box_per_image'],
batch_size=config['train']['batch_size'],
min_net_size=config['model']['min_input_size'],
max_net_size=config['model']['max_input_size'],
shuffle=True,
jitter=0.0,
norm=normalize)
###############################
# Load the model and do evaluation
###############################
infer_model = load_model(config['train']['saved_weights_name'])
# compute mAP for all the classes
average_precisions = evaluate(infer_model, valid_generator)
# print the score
for label, average_precision in average_precisions.items():
print(labels[label] + ': {:.4f}'.format(average_precision))
print('mAP: {:.4f}'.format(
sum(average_precisions.values()) / len(average_precisions)))
def test(args):
config_path = 'config.json'
with open(config_path) as config_buffer:
config = json.loads(config_buffer.read())
data_dir = args.data_dir
annotation_dir = args.annotation_dir
batch_size = args.batch_size
images_path = data_dir
annotation_path = annotation_dir
labels = config['model']['labels']
# images_path = config['train']['train_image_folder']
# annotation_path = config['train']['train_annot_folder']
# batch_size=config['train']['batch_size']
test_list, label_list = parse_voc(images_path, annotation_path, labels)
label_list = labels
max_box_per_image = max([len(images['object']) for images in test_list])
test_generator = Dataloader(train_list=test_list,
label_list=label_list,
anchors=config['model']['anchors'],
max_box_per_image=max_box_per_image,
batch_size=batch_size)
json_file = open('logs/models/model_architecture.json', 'r')
infer_model_json = json_file.read()
json_file.close()
test_model = model_from_json(infer_model_json)
test_model.load_weights('./logs/weights/voc.h5')
average_precisions = evaluate(test_model, test_generator)
print('\n')
for label, average_precision in average_precisions.items():
print(labels[label] + ': {:.4f}'.format(average_precision))
print('mAP: {:.4f}'.format(
sum(average_precisions.values()) / len(average_precisions)))
'npratio': 4,
'dropout_p': 0.2,
'query_dim': 200,
'embedding_dim': 300,
'filter_num': 400,
'value_dim': 16,
'head_num': 16,
'epochs': 5,
'metrics': 'group_auc,ndcg@5,ndcg@10,mean_mrr',
'attrs': ['title'],
}
hparams = load_hparams(hparams)
device = torch.device(hparams['device'])
vocab, loader_train, loader_test, loader_validate = prepare(hparams,
validate=True)
gcaModel = GCAModel(vocab=vocab, hparams=hparams).to(device)
if hparams['mode'] == 'test':
gcaModel.load_state_dict(torch.load(hparams['save_path']))
print("testing...")
evaluate(gcaModel, hparams, loader_test)
elif hparams['mode'] == 'train':
train(gcaModel,
hparams,
loader_train,
loader_test,
loader_validate,
tb=True)
def _main_(args):
config_path = args.conf
with open(config_path) as config_buffer:
config = json.loads(config_buffer.read())
###############################
# Create the validation generator
###############################
valid_ints, labels = parse_voc_annotation(
config['valid']['valid_annot_folder'],
config['valid']['valid_image_folder'], config['valid']['cache_name'],
config['model']['labels'])
labels = labels.keys() if len(
config['model']['labels']) == 0 else config['model']['labels']
labels = sorted(labels)
valid_generator = BatchGenerator(
instances=valid_ints,
anchors=config['model']['anchors'],
labels=labels,
downsample=
32, # ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image=0,
batch_size=config['train']['batch_size'],
min_net_size=config['model']['min_input_size'],
max_net_size=config['model']['max_input_size'],
shuffle=True,
jitter=0.0,
norm=normalize)
###############################
# Load the model and do evaluation
###############################
###############################
# Create the model
##############################
train_model, infer_model = create_yolov3_model(
nb_class=len(labels),
anchors=config['model']['anchors'],
max_box_per_image=0,
max_grid=[
config['model']['max_input_size'],
config['model']['max_input_size']
],
batch_size=config['train']['batch_size'],
warmup_batches=0,
ignore_thresh=config['train']['ignore_thresh'],
grid_scales=config['train']['grid_scales'],
obj_scale=config['train']['obj_scale'],
noobj_scale=config['train']['noobj_scale'],
xywh_scale=config['train']['xywh_scale'],
class_scale=config['train']['class_scale'],
)
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
saved_weights_name = config['train']['saved_weights_name']
lr = config['train']['learning_rate'],
infer_model.load_weights(saved_weights_name)
optimizer = Adam(lr=lr, clipnorm=0.001)
infer_model.compile(loss=dummy_loss, optimizer=optimizer)
infer_model.summary()
#infer_model = load_model(config['train']['saved_weights_name'])
# compute mAP for all the classes
recall, precision, average_precisions = evaluate(infer_model,
valid_generator)
# print the score
for label, average_precision in average_precisions.items():
print(labels[label] + ': {:.4f}'.format(average_precision))
print('mAP: {:.4f}'.format(
sum(average_precisions.values()) / len(average_precisions)))
return recall, precision, average_precisions
def _main_(args):
config_path = args.conf
with open(config_path) as config_buffer:
config = json.loads(config_buffer.read())
###############################
# Parse the annotations
###############################
train_ints, valid_ints, labels, max_box_per_image = create_training_instances(
config['train']['train_annot_folder'],
config['train']['train_image_folder'], config['train']['cache_name'],
config['valid']['valid_annot_folder'],
config['valid']['valid_image_folder'], config['valid']['cache_name'],
config['model']['labels'])
print('\nTraining on: \t' + str(labels) + '\n')
###############################
# Create the generators
###############################
train_generator = BatchGenerator(
instances=train_ints,
anchors=config['model']['anchors'],
labels=labels,
downsample=
32, # ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image=max_box_per_image,
batch_size=config['train']['batch_size'],
min_net_size=config['model']['min_input_size'],
max_net_size=config['model']['max_input_size'],
shuffle=True,
jitter=0.3,
norm=normalize)
valid_generator = BatchGenerator(
instances=valid_ints,
anchors=config['model']['anchors'],
labels=labels,
downsample=
32, # ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image=max_box_per_image,
batch_size=config['train']['batch_size'],
min_net_size=config['model']['min_input_size'],
max_net_size=config['model']['max_input_size'],
shuffle=True,
jitter=0.0,
norm=normalize)
###############################
# Create the model
###############################
if os.path.exists(config['train']['saved_weights_name']):
config['train']['warmup_epochs'] = 0
warmup_batches = config['train']['warmup_epochs'] * (
config['train']['train_times'] * len(train_generator))
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
multi_gpu = len(config['train']['gpus'].split(','))
train_model, infer_model = create_model(
nb_class=len(labels),
anchors=config['model']['anchors'],
max_box_per_image=max_box_per_image,
max_grid=[
config['model']['max_input_size'],
config['model']['max_input_size']
],
batch_size=config['train']['batch_size'],
warmup_batches=warmup_batches,
ignore_thresh=config['train']['ignore_thresh'],
multi_gpu=multi_gpu,
saved_weights_name=config['train']['saved_weights_name'],
lr=config['train']['learning_rate'],
grid_scales=config['train']['grid_scales'],
obj_scale=config['train']['obj_scale'],
noobj_scale=config['train']['noobj_scale'],
xywh_scale=config['train']['xywh_scale'],
class_scale=config['train']['class_scale'],
)
###############################
# Kick off the training
###############################
callbacks = create_callbacks(config['train']['saved_weights_name'],
config['train']['tensorboard_dir'],
infer_model)
history = train_model.fit_generator(
generator=train_generator,
steps_per_epoch=len(train_generator) *
config['train']['train_times'], #(train_images/4)*train_times
epochs=config['train']['nb_epochs'] + config['train']['warmup_epochs'],
verbose=2 if config['train']['debug'] else 1,
callbacks=callbacks,
workers=4,
max_queue_size=8,
validation_data=valid_generator,
validation_steps=len(valid_generator) * config['valid']['valid_times'])
# Prepare folder to save in
time = datetime.now()
time_str = time.strftime("%Y%m%d-%H:%M")
cwd = os.getcwd()
save_path = os.path.join(cwd, time_str)
print('Data saved in: ' + save_path)
makedirs(save_path)
# Copy and save config file
save_config_path = os.path.join(save_path, 'config.json')
shutil.copy2(config_path, save_config_path)
# plotting train data and saving mat file with data
plot_matlab(history, config['data']['plot_png'],
config['data']['save_mat'], save_path)
# Save the loaded models to .TXT
save_template_path = os.path.join(save_path, 'train_model_params.txt')
with open(save_template_path, 'w') as fh:
# Pass the file handle in as a lambda function to make it callable
train_model.summary(print_fn=lambda x: fh.write(x + '\n'))
# make a GPU version of infer_model for evaluation
if multi_gpu > 1:
infer_model = load_model(config['train']['saved_weights_name'])
###############################
# Run the evaluation
###############################
# compute mAP for all the classes
average_precisions = evaluate(infer_model, valid_generator)
# print the score
print('========== VALIDATION ==========')
for label, average_precision in average_precisions.items():
print(labels[label] + ': {:.4f}'.format(average_precision))
print('mAP: {:.4f}'.format(
sum(average_precisions.values()) / len(average_precisions)))
def _main_(args):
config_path = args.conf
with open(config_path) as config_buffer:
config = json.loads(config_buffer.read())
###############################
# Parse the annotations
###############################
train_ints, valid_ints, labels = create_training_instances(
config['train']['train_annot_folder'],
config['train']['train_image_folder'], config['train']['cache_name'],
config['valid']['valid_annot_folder'],
config['valid']['valid_image_folder'], config['valid']['cache_name'],
config['model']['labels'])
###############################
# Create the generators
###############################
train_generator = BatchGenerator(
instances=train_ints,
anchors=config['model']['anchors'],
labels=labels,
downsample=
32, # ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image=config['model']['max_box_per_image'],
batch_size=config['train']['batch_size'],
min_net_size=config['model']['min_input_size'],
max_net_size=config['model']['max_input_size'],
shuffle=True,
jitter=0.3,
norm=normalize)
valid_generator = BatchGenerator(
instances=valid_ints,
anchors=config['model']['anchors'],
labels=labels,
downsample=
32, # ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image=config['model']['max_box_per_image'],
batch_size=config['train']['batch_size'],
min_net_size=config['model']['min_input_size'],
max_net_size=config['model']['max_input_size'],
shuffle=True,
jitter=0.0,
norm=normalize)
###############################
# Create the model
###############################
if os.path.exists(config['train']['saved_weights_name']):
warmup_batches = 0 # no need warmup if the pretrained weight exists
else:
warmup_batches = config['train']['warmup_epochs'] * (config['train']['train_times']*len(train_generator) + \
config['valid']['valid_times']*len(valid_generator))
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
multi_gpu = len(config['train']['gpus'].split(','))
train_model, infer_model = create_model(
nb_class=len(labels),
anchors=config['model']['anchors'],
max_box_per_image=config['model']['max_box_per_image'],
max_grid=[
config['model']['max_input_size'],
config['model']['max_input_size']
],
batch_size=config['train']['batch_size'],
warmup_batches=warmup_batches,
ignore_thresh=config['train']['ignore_thresh'],
multi_gpu=multi_gpu,
saved_weights_name=config['train']['saved_weights_name'])
###############################
# Kick off the training
###############################
optimizer = Adam(lr=config['train']['learning_rate'],
beta_1=0.9,
beta_2=0.999,
epsilon=1e-08,
decay=0.0)
train_model.compile(loss=dummy_loss, optimizer=optimizer)
callbacks = create_callbacks(config['train']['saved_weights_name'])
try:
train_model.fit_generator(generator=train_generator,
steps_per_epoch=len(train_generator) *
config['train']['train_times'],
epochs=config['train']['nb_epochs'] +
config['train']['warmup_epochs'],
verbose=2 if config['train']['debug'] else 1,
validation_data=valid_generator,
validation_steps=len(valid_generator) *
config['valid']['valid_times'],
callbacks=callbacks,
workers=4,
max_queue_size=8)
except:
pass
# load the best weight before early stop
train_model.load_weights(config['train']['saved_weights_name'])
if multi_gpu > 1:
# fix the saved model structure when multi_gpu > 1
train_model.get_layer("model_1").save(
config['train']['saved_weights_name'])
# load the best weight to the infer_model
infer_model.load_weights(config['train']['saved_weights_name'])
# save the weight with the model structure of infer_model
infer_model.save(config['train']['saved_weights_name'])
# make a GPU version of infer_model for evaluation
if multi_gpu > 1:
infer_model = load_model(config['train']['saved_weights_name'])
###############################
# Run the evaluation
###############################
# compute mAP for all the classes
average_precisions = evaluate(infer_model, valid_generator)
# print the score
for label, average_precision in average_precisions.items():
print(labels[label] + ': {:.4f}'.format(average_precision))
print('mAP: {:.4f}'.format(
sum(average_precisions.values()) / len(average_precisions)))
'depth_loss_function': depth_loss_function
}
# Load model into GPU / CPU
print('Loading model...')
model = load_model(args.model, custom_objects=custom_objects, compile=False)
# Load test data
print('Loading test data...', end='')
import numpy as np
from data import extract_zip
data = extract_zip('nyu_test.zip')
from io import BytesIO
rgb = np.load(BytesIO(data['eigen_test_rgb.npy']))
depth = np.load(BytesIO(data['eigen_test_depth.npy']))
crop = np.load(BytesIO(data['eigen_test_crop.npy']))
print('Test data loaded.\n')
start = time.time()
print('Testing...')
e = evaluate(model, rgb, depth, crop, batch_size=6)
print("{:>10}, {:>10}, {:>10}, {:>10}, {:>10}, {:>10}".format(
'a1', 'a2', 'a3', 'rel', 'rms', 'log_10'))
print("{:10.4f}, {:10.4f}, {:10.4f}, {:10.4f}, {:10.4f}, {:10.4f}".format(
e[0], e[1], e[2], e[3], e[4], e[5]))
end = time.time()
print('\nTest time', end - start, 's')
def main():
'''
Training and evaluation of the model.
'''
print('Training starts...')
for epoch in range(num_of_epoch):
print('\nEpoch', epoch + 1)
# log the start time of the epoch
start = time.time()
# set the models in training mode
clstm.train()
policy_s.train()
policy_n.train()
policy_c.train()
# reset the count of reread_or_skim_times
reread_or_skim_times = 0
policy_loss_sum = []
encoder_loss_sum = []
baseline_value_batch = []
for index, train in enumerate(train_iterator):
label = train.label.to(
torch.long
) # for cross entropy loss, the long type is required
text = train.text.view(CHUNCK_SIZE, BATCH_SIZE,
CHUNCK_SIZE) # transform 1*400 to 20*1*20
curr_step = 0 # the position of the current chunk
h_0 = torch.zeros([1, 1, 128]).to(device) # run on GPU
c_0 = torch.zeros([1, 1, 128]).to(device)
count = 0 # maximum skim/reread time: 5
baseline_value_ep = []
saved_log_probs = [] # for the use of policy gradient update
# collect the computational costs for every time step
cost_ep = []
while curr_step < CHUNCK_SIZE and count < 5:
# Loop until a text can be classified or currstep is up to 20 or count reach the maximum i.e. 5.
# update count
count += 1
# pass the input through cnn-lstm and policy s
text_input = text[curr_step] # text_input 1*20
ht, ct = clstm(text_input, h_0, c_0) # 1 * 128
# separate the value which is the input of value net
ht_ = ht.clone().detach().requires_grad_(True)
# compute a baseline value for the value network
bi = value_net(ht_)
# 1 * 1 * 128, next input of lstm
h_0 = ht.unsqueeze(0)
c_0 = ct
# draw a stop decision
stop_decision, log_prob_s = sample_policy_s(ht, policy_s)
stop_decision = stop_decision.item()
if stop_decision == 1: # classify
break
else:
reread_or_skim_times += 1
# draw an action (reread or skip)
step, log_prob_n = sample_policy_n(ht, policy_n)
curr_step += int(step) # reread or skip
if curr_step < CHUNCK_SIZE and count < 5:
# If the code can still execute the next loop, it is not the last time step.
cost_ep.append(clstm_cost + s_cost + n_cost)
# add the baseline value
baseline_value_ep.append(bi)
# add the log prob for the current actions
saved_log_probs.append(log_prob_s + log_prob_n)
# draw a predicted label
output_c = policy_c(ht)
# cross entrpy loss input shape: input(N, C), target(N)
loss = criterion(output_c, label) # positive value
# draw a predicted label
pred_label, log_prob_c = sample_policy_c(output_c)
if stop_decision == 1:
# add the cost of the last time step
cost_ep.append(clstm_cost + s_cost + c_cost)
saved_log_probs.append(log_prob_s + log_prob_c)
else:
# add the cost of the last time step
cost_ep.append(clstm_cost + s_cost + c_cost + n_cost)
# At the moment, the probability of drawing a stop decision is 1,
# so its log probability is zero which can be ignored in th sum.
saved_log_probs.append(log_prob_c.unsqueeze(0))
# add the baseline value
baseline_value_ep.append(bi)
# add the cross entropy loss
encoder_loss_sum.append(loss)
# compute the policy losses and value losses for the current episode
policy_loss_ep, value_losses = compute_policy_value_losses(
cost_ep, loss, saved_log_probs, baseline_value_ep, alpha,
gamma)
policy_loss_sum.append(torch.cat(policy_loss_ep).sum())
baseline_value_batch.append(torch.cat(value_losses).sum())
# update gradients
if (index + 1
) % batch_sz == 0: # take the average of samples, backprop
finish_episode(policy_loss_sum, encoder_loss_sum,
baseline_value_batch)
del policy_loss_sum[:], encoder_loss_sum[:], baseline_value_batch[:]
if (index + 1) % 2000 == 0:
print(f'\n current episode: {index + 1}')
# log the current position of the text which the agent has gone through
print('curr_step: ', curr_step)
# log the sum of the rereading and skimming times
print(f'current reread_or_skim_times: {reread_or_skim_times}')
print('Epoch time elapsed: %.2f s' % (time.time() - start))
print('reread_or_skim_times in this epoch:', reread_or_skim_times)
count_all, count_correct = evaluate(clstm, policy_s, policy_n,
policy_c, valid_iterator)
print('Epoch: %s, Accuracy on the validation set: %.2f' %
(epoch + 1, count_correct / count_all))
count_all, count_correct = evaluate(clstm, policy_s, policy_n,
policy_c, train_iterator)
print('Epoch: %s, Accuracy on the training set: %.2f' %
(epoch + 1, count_correct / count_all))
print('Compute the accuracy on the testing set...')
count_all, count_correct = evaluate(clstm, policy_s, policy_n, policy_c,
test_iterator)
print('Accuracy on the testing set: %.2f' % (count_correct / count_all))
def _main_(args):
config_path = args.conf
with open(config_path) as config_buffer:
config = json.loads(config_buffer.read())
###############################
# Parse the annotations
###############################
train_ints, valid_ints, labels, max_box_per_image = create_training_instances(
config['train']['train_annot_folder'],
config['train']['train_image_folder'],
config['train']['cache_name'],
config['valid']['valid_annot_folder'],
config['valid']['valid_image_folder'],
config['valid']['cache_name'],
config['model']['labels']
)
print('\nTraining on: \t' + str(labels) + '\n')
###############################
# Create the generators
###############################
train_generator = BatchGenerator(
instances = train_ints,
anchors = config['model']['anchors'],
labels = labels,
downsample = 32, # ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image = max_box_per_image,
batch_size = config['train']['batch_size'],
min_net_size = config['model']['min_input_size'],
max_net_size = config['model']['max_input_size'],
shuffle = True,
jitter = 0.3,
norm = normalize
)
valid_generator = BatchGenerator(
instances = valid_ints,
anchors = config['model']['anchors'],
labels = labels,
downsample = 32, # ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image = max_box_per_image,
batch_size = config['train']['batch_size'],
min_net_size = config['model']['min_input_size'],
max_net_size = config['model']['max_input_size'],
shuffle = True,
jitter = 0.0,
norm = normalize
)
###############################
# Create the model
###############################
if os.path.exists(config['train']['saved_weights_name']):
config['train']['warmup_epochs'] = 0
warmup_batches = config['train']['warmup_epochs'] * (config['train']['train_times']*len(train_generator))
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
multi_gpu = len(config['train']['gpus'].split(','))
train_model, infer_model = create_model(
nb_class = len(labels),
anchors = config['model']['anchors'],
max_box_per_image = max_box_per_image,
max_grid = [config['model']['max_input_size'], config['model']['max_input_size']],
batch_size = config['train']['batch_size'],
warmup_batches = warmup_batches,
ignore_thresh = config['train']['ignore_thresh'],
multi_gpu = multi_gpu,
saved_weights_name = config['train']['saved_weights_name'],
lr = config['train']['learning_rate'],
grid_scales = config['train']['grid_scales'],
obj_scale = config['train']['obj_scale'],
noobj_scale = config['train']['noobj_scale'],
xywh_scale = config['train']['xywh_scale'],
class_scale = config['train']['class_scale'],
)
###############################
# Kick off the training
###############################
callbacks = create_callbacks(config['train']['saved_weights_name'], config['train']['tensorboard_dir'], infer_model)
train_model.fit_generator(
generator = train_generator,
steps_per_epoch = len(train_generator) * config['train']['train_times'],
epochs = config['train']['nb_epochs'] + config['train']['warmup_epochs'],
verbose = 2 if config['train']['debug'] else 1,
callbacks = callbacks,
workers = 4,
max_queue_size = 8
)
# make a GPU version of infer_model for evaluation
if multi_gpu > 1:
infer_model = load_model(config['train']['saved_weights_name'])
###############################
# Run the evaluation
###############################
# compute mAP for all the classes
average_precisions = evaluate(infer_model, valid_generator)
# print the score
for label, average_precision in average_precisions.items():
print(labels[label] + ': {:.4f}'.format(average_precision))
print('mAP: {:.4f}'.format(sum(average_precisions.values()) / len(average_precisions)))
def on_epoch_end(self, epoch, logs=None):
if not test_set == None:
# Samples using current model
import matplotlib.pyplot as plt
from skimage.transform import resize
plasma = plt.get_cmap('plasma')
minDepth, maxDepth = 10, 1000
train_samples = []
test_samples = []
for i in range(self.num_samples):
x_train, y_train = train_generator.__getitem__(
self.train_idx[i], False)
x_test, y_test = test_generator[self.test_idx[i]]
x_train, y_train = x_train[0], np.clip(
DepthNorm(y_train[0], maxDepth=1000), minDepth,
maxDepth) / maxDepth
x_test, y_test = x_test[0], np.clip(
DepthNorm(y_test[0], maxDepth=1000), minDepth,
maxDepth) / maxDepth
h, w = y_train.shape[0], y_train.shape[1]
rgb_train = resize(x_train, (h, w),
preserve_range=True,
mode='reflect',
anti_aliasing=True)
rgb_test = resize(x_test, (h, w),
preserve_range=True,
mode='reflect',
anti_aliasing=True)
gt_train = plasma(y_train[:, :, 0])[:, :, :3]
gt_test = plasma(y_test[:, :, 0])[:, :, :3]
predict_train = plasma(
predict(model,
x_train,
minDepth=minDepth,
maxDepth=maxDepth)[0, :, :, 0])[:, :, :3]
predict_test = plasma(
predict(model,
x_test,
minDepth=minDepth,
maxDepth=maxDepth)[0, :, :, 0])[:, :, :3]
train_samples.append(
np.vstack([rgb_train, gt_train, predict_train]))
test_samples.append(
np.vstack([rgb_test, gt_test, predict_test]))
self.writer.add_summary(
tf.Summary(value=[
tf.Summary.Value(tag='Train',
image=make_image(
255 * np.hstack(train_samples)))
]), epoch)
self.writer.add_summary(
tf.Summary(value=[
tf.Summary.Value(tag='Test',
image=make_image(
255 * np.hstack(test_samples)))
]), epoch)
# Metrics
e = evaluate(model,
test_set['rgb'],
test_set['depth'],
test_set['crop'],
batch_size=6,
verbose=True)
logs.update({'rel': e[3]})
logs.update({'rms': e[4]})
logs.update({'log10': e[5]})
super().on_epoch_end(epoch, logs)
hparams = {
'name': 'knrm',
'batch_size': 100,
'title_size': 20,
'his_size': 50,
'npratio': 4,
'embedding_dim': 300,
'kernel_num': 11,
'metrics': 'group_auc,ndcg@5,ndcg@10,mean_mrr',
'attrs': ['title'],
}
hparams = load_hparams(hparams)
device = torch.device(hparams['device'])
vocab, loader_train, loader_test, loader_validate = prepare(hparams,
validate=True)
knrmModel = KNRMModel(vocab=vocab, hparams=hparams).to(device)
if hparams['mode'] == 'test':
knrmModel.load_state_dict(torch.load(hparams['save_path']))
print("testing...")
evaluate(knrmModel, hparams, loader_test)
elif hparams['mode'] == 'train':
train(knrmModel,
hparams,
loader_train,
loader_test,
loader_validate,
tb=True)
def _main():
config_path = './local_config.json'
LABELS = read_category()
with open(config_path) as config_buffer:
config = json.loads(config_buffer.read())
''' Parse annotations '''
config['model']['labels'] = LABELS
train_ints, valid_ints, labels, max_box_per_image = create_train_valid_set(
config['train']['train_annot_folder'],
config['train']['train_image_folder'],
config['valid']['valid_annot_folder'],
config['valid']['valid_image_folder'], config['model']['labels'])
''' Create generators '''
# check if images are normal after BatchGenerator
batches = BatchGenerator(instances=np.append(train_ints, valid_ints),
anchors=config['model']['anchors'],
labels=labels,
downsample=32,
max_box_per_image=max_box_per_image,
batch_size=config['train']['batch_size'],
shuffle=False,
jitter=False)
img = batches[0][0][0][5]
# plt.imshow(img.astype('uint8'))
train_generator = BatchGenerator(
instances=train_ints,
anchors=config['model']['anchors'],
labels=labels,
downsample=
32, # ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image=max_box_per_image,
batch_size=config['train']['batch_size'],
# min_net_size=config['model']['min_input_size'],
# max_net_size=config['model']['max_input_size'],
shuffle=True,
jitter=True, # add 10% noise for each image for training
norm=normalize)
img = train_generator[0][0][0][5]
# plt.imshow(img.astype('float'))
valid_generator = BatchGenerator(
instances=valid_ints,
anchors=config['model']['anchors'],
labels=labels,
downsample=
32, # ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image=max_box_per_image,
batch_size=config['train']['batch_size'],
# min_net_size=config['model']['min_input_size'],
# max_net_size=config['model']['max_input_size'],
shuffle=True,
jitter=False,
norm=normalize)
''' Create the model '''
if os.path.exists(config['train']['saved_weights_name']):
config['train']['warmup_epochs'] = 0
warmup_batches = config['train']['warmup_epochs'] * (
config['train']['train_times'] * len(train_generator))
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
multi_gpu = len(config['train']['gpus'].split(','))
train_model, infer_model = create_model(
nb_class=len(labels),
anchors=config['model']['anchors'],
max_box_per_image=max_box_per_image,
max_grid=[
config['model']['max_input_size'],
config['model']['max_input_size']
],
batch_size=config['train']['batch_size'],
warmup_batches=warmup_batches,
ignore_thresh=config['train']['ignore_thresh'],
multi_gpu=multi_gpu,
saved_weights_name=config['train']['saved_weights_name'],
lr=config['train']['learning_rate'],
grid_scales=config['train']['grid_scales'],
obj_scale=config['train']['obj_scale'],
noobj_scale=config['train']['noobj_scale'],
xywh_scale=config['train']['xywh_scale'],
class_scale=config['train']['class_scale'],
)
print('\ntrain_model: \n')
print(train_model.summary())
print('\ninfer_model: \n')
print(infer_model.summary())
''' Kick off the training '''
callbacks = create_callbacks(config['train']['saved_weights_name'],
config['train']['tensorboard_dir'],
infer_model)
train_model.fit_generator(
generator=train_generator,
steps_per_epoch=len(
train_generator), # * config['train']['train_times'],
epochs=config['train']['nb_epochs'] + config['train']['warmup_epochs'],
verbose=2 if config['train']['debug'] else 1,
callbacks=callbacks,
workers=16,
max_queue_size=8)
# make a GPU version of infer_model for evaluation
if multi_gpu > 1:
infer_model = load_model(config['train']['saved_weights_name'])
''' Evaluate the result '''
# compute mAP for all the classes
average_precisions = evaluate(infer_model, valid_generator)
# print the score
for label, average_precision in average_precisions.items():
print(labels[label] + ': {:.4f}'.format(average_precision))
print('mAP: {:.4f}'.format(
sum(average_precisions.values()) / len(average_precisions)))
callbacks = callbacks,
workers = 4,
max_queue_size = 8
)
print("[INFO] Saving Model...")
print("[INFO] Start Evalutating Model...")
# make a GPU version of infer_model for evaluation
if multi_gpu > 1:
infer_model = load_model(config['train']['saved_weights_name'])
###############################
# Run the evaluation
###############################
# compute mAP for all the classes
average_precisions = evaluate(infer_model, valid_generator)
ap = []
# print the mAP score
for label, average_precision in average_precisions.items():
print(labels[label] + ' average precision(AP): {:.6f}'.format(average_precision['ap']))
ap.append(average_precision['ap'])
print(labels[label] + ' recall: {:.6f}'.format(average_precision['recall']))
print(labels[label] + ' precision: {:.6f}'.format(average_precision['precision']))
print('[INFO] mAP: {:.6f}'.format(sum(ap) / len(ap)))
print("[INFO] Completed...")
if __name__ == '__main__':
argparser = argparse.ArgumentParser(description='train and evaluate YOLO_v3 model on any dataset')
argparser.add_argument('-c', '--conf', help='path to configuration file')
def main(_):
image_path = FLAGS.test
csv_path = os.path.splitext(image_path)[0] + ".csv"
# ---------- create model ----------------#
x = tf.placeholder("float", shape=[None, FLAGS.input_size * FLAGS.input_size])
keep_prob = tf.placeholder("float")
global_step = tf.Variable(0, trainable=False, name='global_step')
model = nn.create(x, keep_prob)
# ---------- restore model ---------------#
saver = tf.train.Saver()
if tf.train.latest_checkpoint(FLAGS.checkpoint_dir) != None:
saver.restore(sess, tf.train.latest_checkpoint(FLAGS.checkpoint_dir))
# ---------- object detection ------------#
print 'starting detection of ' + FLAGS.test + '...'
img = utils.getImage(image_path)
img = cv2.copyMakeBorder(img, FLAGS.max_window_size, FLAGS.max_window_size, FLAGS.max_window_size, FLAGS.max_window_size, cv2.BORDER_REPLICATE)
start = time.time()
#sliding window detection
mask = create_mask(model, x, keep_prob, img)
elapsed = time.time() - start
print 'detection time: %d' % (elapsed)
# ------------- evaluation --------------#
global_step = tf.train.global_step(sess, global_step)
ground_truth_data = utils.get_ground_truth_data(csv_path)
ground_truth_data = [(x + FLAGS.max_window_size,y + FLAGS.max_window_size) for (x,y) in ground_truth_data]
for th in [150, 160, 170, 180, 190, 200, 210, 220, 230, 240, 250]:
detected = mask_to_objects(mask, th)
tp, fn, fp = utils.evaluate(ground_truth_data, detected)
# ----------------output ----------------#
# image output
"""
img_out = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB) * 255
for (r,score) in candidates:
cv2.rectangle(img_out, (r.x,r.y), (r.x2(), r.y2()), [200,200,200], 2)
for r in detected:
cv2.rectangle(img_out, (r.x,r.y), (r.x2(), r.y2()), [0,255,0], 2)
for c in ground_truth_data:
cv2.circle(img_out, (c[0], c[1]), 3, [0,0,255],3)
output_file = "out" + '_' + str(global_step) + 'its_' + str(FLAGS.step_size) + 'step_' + str(th) + 'threshold_' + str(datetime.datetime.now())
cv2.imwrite(FLAGS.output_dir + output_file + '.png', img_out)
"""
# csv output
with open(FLAGS.output_dir + 'results.csv', 'ab') as file:
writer = csv.writer(file, delimiter=',')
writer.writerow([FLAGS.test, str(elapsed),
str(global_step), str(len(ground_truth_data)), str(th),
str(len(detected)), str(FLAGS.step_size), str(tp), str(fp), str(fn)])
def eval_map(self):
return evaluate(self.model, self.data_generator)
def _main_(args):
config_path = args.conf
with open(config_path) as config_buffer:
config = json.loads(config_buffer.read())
###############################
# Parse the annotations
###############################
train_ints, valid_ints, labels, max_box_per_image = create_training_instances(
config['train']['train_annot_folder'],
config['train']['train_image_folder'], config['train']['cache_name'],
config['valid']['valid_annot_folder'],
config['valid']['valid_image_folder'], config['valid']['cache_name'],
config['model']['labels'])
print('\nTraining on: \t' + str(labels) + '\n')
###############################
# Create the generators
###############################
train_generator = BatchGenerator(
instances=train_ints,
anchors=config['model']['anchors'],
labels=labels,
downsample=
32, # ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image=max_box_per_image,
batch_size=config['train']['batch_size'],
min_net_size=config['model']['min_input_size'],
max_net_size=config['model']['max_input_size'],
shuffle=True,
jitter=0.3,
norm=normalize)
valid_generator = BatchGenerator(
instances=valid_ints,
anchors=config['model']['anchors'],
labels=labels,
downsample=
32, # ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image=max_box_per_image,
batch_size=config['train']['batch_size'],
min_net_size=config['model']['min_input_size'],
max_net_size=config['model']['max_input_size'],
shuffle=True,
jitter=0.0,
norm=normalize)
###############################
# Create the model
###############################
if os.path.exists(config['train']['saved_weights_name']):
config['train']['warmup_epochs'] = 0
warmup_batches = config['train']['warmup_epochs'] * (
config['train']['train_times'] * len(train_generator))
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
multi_gpu = len(config['train']['gpus'].split(','))
print('multi_gpu:' + str(multi_gpu))
train_model, infer_model = create_model(
nb_class=len(labels),
anchors=config['model']['anchors'],
max_box_per_image=max_box_per_image,
max_grid=[
config['model']['max_input_size'],
config['model']['max_input_size']
],
batch_size=config['train']['batch_size'],
warmup_batches=warmup_batches,
ignore_thresh=config['train']['ignore_thresh'],
multi_gpu=multi_gpu,
saved_weights_name=config['train']['saved_weights_name'],
lr=config['train']['learning_rate'],
grid_scales=config['train']['grid_scales'],
obj_scale=config['train']['obj_scale'],
noobj_scale=config['train']['noobj_scale'],
xywh_scale=config['train']['xywh_scale'],
class_scale=config['train']['class_scale'],
backend=config['model']['backend'])
###############################
# Kick off the training
###############################
callbacks = create_callbacks(config['train']['saved_weights_name'],
config['train']['tensorboard_dir'],
infer_model)
train_model.fit(
x=train_generator,
validation_data=valid_generator,
steps_per_epoch=len(train_generator) * config['train']['train_times'],
epochs=config['train']['nb_epochs'] + config['train']['warmup_epochs'],
verbose=2 if config['train']['debug'] else 1,
workers=4,
max_queue_size=8,
callbacks=callbacks)
train_model.load_weights(config['train']['saved_weights_name'])
infer_model.save(config['train']['saved_weights_name'])
###############################
# Run the evaluation
###############################
# compute mAP for all the classes
average_precisions = evaluate(infer_model, valid_generator)
# print the score
total_instances = []
precisions = []
for label, (average_precision,
num_annotations) in average_precisions.items():
print('{:.0f} instances of class'.format(num_annotations),
labels[label],
'with average precision: {:.4f}'.format(average_precision))
total_instances.append(num_annotations)
precisions.append(average_precision)
if sum(total_instances) == 0:
print('No test instances found.')
return
print('mAP using the weighted average of precisions among classes: {:.4f}'.
format(
sum([a * b for a, b in zip(total_instances, precisions)]) /
sum(total_instances)))
print('mAP: {:.4f}'.format(
sum(precisions) / sum(x > 0 for x in total_instances)))
def _main_(args):
config_path = args.conf
with open(config_path) as config_buffer:
config = json.loads(config_buffer.read())
###############################
# Parse the annotations
###############################
train_ints, valid_ints, labels, max_box_per_image = create_training_instances(
config['train']['train_annot_folder'],
config['train']['train_image_folder'], config['train']['cache_name'],
config['valid']['valid_annot_folder'],
config['valid']['valid_image_folder'], config['valid']['cache_name'],
config['model']['labels'])
print('\nTraining on: \t' + str(labels) + '\n')
###############################
# Create the generators
###############################
train_generator = BatchGenerator(
instances=train_ints,
anchors=config['model']['anchors'],
labels=labels,
downsample=
32, # ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image=max_box_per_image,
batch_size=config['train']['batch_size'],
min_net_size=config['model']['min_input_size'],
max_net_size=config['model']['max_input_size'],
shuffle=True,
jitter=0.3,
norm=normalize)
valid_generator = BatchGenerator(
instances=valid_ints,
anchors=config['model']['anchors'],
labels=labels,
downsample=
32, # ratio between network input's size and network output's size, 32 for YOLOv3
max_box_per_image=max_box_per_image,
batch_size=config['train']['batch_size'],
min_net_size=config['model']['min_input_size'],
max_net_size=config['model']['max_input_size'],
shuffle=True,
jitter=0.0,
norm=normalize)
###############################
# Create the model
###############################
if os.path.exists(config['train']['saved_weights_name']):
config['train']['warmup_epochs'] = 0
warmup_batches = config['train']['warmup_epochs'] * (
config['train']['train_times'] * len(train_generator))
os.environ['CUDA_VISIBLE_DEVICES'] = config['train']['gpus']
multi_gpu = len(config['train']['gpus'].split(','))
train_model, infer_model = create_model(
nb_class=len(labels),
anchors=config['model']['anchors'],
max_box_per_image=max_box_per_image,
max_grid=[
config['model']['max_input_size'],
config['model']['max_input_size']
],
batch_size=config['train']['batch_size'],
warmup_batches=warmup_batches,
ignore_thresh=config['train']['ignore_thresh'],
multi_gpu=multi_gpu,
saved_weights_name=config['train']['saved_weights_name'],
lr=config['train']['learning_rate'],
grid_scales=config['train']['grid_scales'],
obj_scale=config['train']['obj_scale'],
noobj_scale=config['train']['noobj_scale'],
xywh_scale=config['train']['xywh_scale'],
class_scale=config['train']['class_scale'],
)
###############################
# Kick off the training
###############################
callbacks = create_callbacks(config['train']['saved_weights_name'],
config['train']['tensorboard_dir'],
infer_model)
train_model.fit_generator(
generator=train_generator,
steps_per_epoch=len(train_generator) * config['train']['train_times'],
epochs=config['train']['nb_epochs'] + config['train']['warmup_epochs'],
verbose=2 if config['train']['debug'] else 1,
callbacks=callbacks,
workers=4,
max_queue_size=8)
# make a GPU version of infer_model for evaluation
if multi_gpu > 1:
infer_model = load_model(config['train']['saved_weights_name'])
###############################
# Run the evaluation
###############################
# compute mAP for all the classes
average_precisions = evaluate(infer_model, valid_generator)
# print the score
for label, average_precision in average_precisions.items():
print(labels[label] + ': {:.4f}'.format(average_precision))
print('mAP: {:.4f}'.format(
sum(average_precisions.values()) / len(average_precisions)))
def main():
text = 'Description of the program arguments'
parser = argparse.ArgumentParser(description=text)
parser.add_argument("--train_path",
"-t",
help="set path where to find training data")
parser.add_argument("--train_start",
help="start data fraction for train data (default 0)")
parser.add_argument("--train_end",
help="end data fraction for train data (default 1)")
parser.add_argument("--test_path",
"-T",
help="set path where to find test data")
parser.add_argument("--test_start",
help="start data fraction for test data (default 0)")
parser.add_argument("--test_end",
help="end data fraction for test data (default 1)")
parser.add_argument("--embedding_path",
"-e",
help="set path where to find french word embeddings")
parser.add_argument(
"--mode",
"-m",
help=
"set mode:\n - 'prediction' / 'e': predict only \n - 'evaluation' / 'e': predict and "
"evaluate predictions")
parser.add_argument(
"--output_path",
"-o",
help=
"set path where to write predictions (if None nothing will be written)"
)
parser.add_argument(
"--beam",
"-b",
help="set beam search size for cyk algorithm (default 10)")
args = parser.parse_args()
def change_none(x, val):
return val if x is None else x
train_path = args.train_path
train_start = float(change_none(args.train_start, 0))
train_end = float(change_none(args.train_end, 1))
test_path = args.test_path
test_start = float(change_none(args.test_start, 0))
test_end = float(change_none(args.test_end, 1))
embedding_path = args.embedding_path
mode = args.mode
beam = int(change_none(args.beam, 10))
output_path = args.output_path
assert mode in ('prediction', 'evaluation', 'e', 'p'), mode
print("#" * 100 + '\n##')
print('##\t- Build grammar from file: %s' % train_path)
print('##\t- Build oov module from embeddings stored in: %s' %
embedding_path)
print('##\t- Make {} based on cyk (beam: {}) on sentences in file: {}'.
format(
'predictions' if mode in ('prediction', 'p') else 'evaluations',
beam, test_path))
if output_path is not None:
print('##\t- Store predictions in file: %s' % output_path)
else:
print("## Don't save predictions")
print('##\n' + "#" * 100)
train_vocabulary, train_grammar_rules, train_rhs_index, train_unary_dic, train_prob_lexicon = get_train_data(
train_path, train_start, train_end)
if mode in ('prediction', 'p'):
test_sentences = get_to_predict_data(test_path, test_start, test_end)
elif mode in ('evaluation', 'e'):
test_sentences, test_labels = get_to_eval_data(test_path, test_start,
test_end)
else:
raise ValueError(
"Should be 'prediction', 'p', 'evaluation' or 'e', not %s" % mode)
# load French word embeddings
fr_words, embeddings, word_id, id_word = get_embeddings(embedding_path)
# Normalize digits by replacing them with #
DIGITS = re.compile("[0-9]", re.UNICODE)
# considered transformations when looking for in vocabulary words
TRANSFOS = [
lambda w: DIGITS.sub("#", w), lambda w: w.lower(), lambda w: w.upper(),
lambda w: w.title()
]
train_embeddings, train_word_id, voc_id_word = process_embeddings(
word_embeddings=embeddings,
word_id_dic=word_id,
vocabulary=train_vocabulary,
re_rules=[lambda s: DIGITS.sub("#", s)])
def oov_handler(word):
return oov(word,
train_vocabulary,
fr_words,
all_embs=embeddings,
all_word_id_dic=word_id,
voc_embs=train_embeddings,
voc_id_word_dic=voc_id_word,
transformations=TRANSFOS,
k=2)
print("Vocabulary-specific embedding shape is {}".format(
train_embeddings.shape))
if mode in ('evaluation', 'e'):
parsed_str, score, parsed = evaluate(test_sentences,
test_labels,
train_grammar_rules,
train_prob_lexicon,
train_rhs_index,
train_unary_dic,
oov_handler,
p_output=True,
beam=beam,
chrono=True)
elif mode in ('prediction', 'p'):
parsed_str, parsed = predict(test_sentences,
train_grammar_rules,
train_prob_lexicon,
train_rhs_index,
train_unary_dic,
oov_handler,
p_output=True,
beam=beam,
chrono=True)
if output_path is not None:
print("Write predictions in %s..." % output_path, end=' ')
write_in_file(output_path, parsed_str)
print('Done')
os.makedirs(os.path.join(opt.output_dir, "models"), exist_ok=True)
os.makedirs(os.path.join(opt.output_dir, "tensorboard"), exist_ok=True)
writer = SummaryWriter(log_dir=os.path.join(opt.output_dir, "tensorboard"))
# ----------
# Training
# ----------
if opt.train:
train(model_G=model_G,
model_D=model_D,
embedder=encoder,
optimizer_G=optimizer_G,
optimizer_D=optimizer_D,
scheduler_G=scheduler_G,
scheduler_D=scheduler_D,
train_loader=train_dataloader,
val_loader=val_dataloader,
adv_loss=adversarial_loss,
opt=opt,
onehot_encoder=onehot_encoder)
else:
assert opt.model_checkpoint is not None, 'no model checkpoint specified'
print("Loading model from state dict...")
load_model(opt.model_checkpoint, model_G)
print("Model loaded.")
evaluate(model_G=model_G,
embedder=encoder,
test_loader=val_dataloader,
device=device,
p=opt.model_checkpoint)
def on_epoch_end(self, epoch, logs=None):
logs = logs or {}
self.epochs_since_last_save += 1
if self.epochs_since_last_save >= self.period:
self.epochs_since_last_save = 0
filepath = self.filepath.format(epoch=epoch + 1, **logs)
if self.save_best_only:
current = logs.get(self.monitor)
if current is None:
warnings.warn(
'Can save best model only with %s available, '
'skipping.' % (self.monitor), RuntimeWarning)
else:
if self.monitor_op(current, self.best):
if self.verbose > 0:
print(
'\nEpoch %05d: %s improved from %0.5f to %0.5f,'
' saving model to %s' %
(epoch + 1, self.monitor, self.best, current,
filepath))
self.best = current
# print(self.save_weights_only)
if self.save_weights_only:
self.model_to_save.save_weights(filepath,
overwrite=True)
else:
self.model_to_save.save(filepath, overwrite=True)
else:
if self.verbose > 0:
print(
'\nEpoch %05d: %s did not improve from %0.5f' %
(epoch + 1, self.monitor, self.best))
# # normal save
# if self.verbose > 0:
# print('\nEpoch %05d: saving model to %s' % (epoch + 1, filepath))
# if self.save_weights_only:
# self.model_to_save.save_weights(filepath, overwrite=True)
# else:
# self.model_to_save.save(filepath, overwrite=True)
else:
if self.verbose > 0:
print('\nEpoch %05d: saving model to %s' %
(epoch + 1, filepath))
if self.save_weights_only:
self.model_to_save.save_weights(filepath, overwrite=True)
else:
self.model_to_save.save(filepath, overwrite=True)
if self.addtion_save:
if (epoch + 1) % 5 == 0 and (epoch + 1) > 19:
average_precisions = evaluate(self.model_to_save,
self.valid_data)
ap = []
print('[INFO] Epoch: %05d' % (epoch + 1))
# print the mAP score
for label, average_precision in average_precisions.items():
print('\n' + self.labels[label] +
' average precision(AP): {:.6f}'.format(
average_precision['ap']))
ap.append(average_precision['ap'])
print(
self.labels[label] +
' recall: {:.6f}'.format(average_precision['recall']))
print(self.labels[label] + ' precision: {:.6f}'.format(
average_precision['precision']))
mAP = sum(ap) / len(ap)
print('[INFO] mAP: {:.6f}'.format(mAP))
if self.best_mAP < mAP:
print(
'[INFO] Best mAP improve from {:.6f} to {:.6f}'.format(
self.best_mAP, mAP))
self.best_mAP = mAP
self.model_to_save.save(
str(self.addtion_save).split('.')[0] + '_mAP_best.h5',
overwrite=True)
else:
print('[INFO] Best mAP did not improve from {:.6f}'.format(
self.best_mAP))
if (epoch + 1) % 10 == 0:
self.model_to_save.save(str(self.addtion_save).split('.')[0] +
'_%04d.h5' % (epoch + 1),
overwrite=True)
super(CustomModelCheckpoint, self).on_batch_end(epoch, logs)
