def _predict(self,
data_generator_function,
steps_per_epoch,
include_datum=True):
data_generator = data_generator_function(include_datum=True)
enqueuer = GeneratorEnqueuer(data_generator, pickle_safe=False)
enqueuer.start(workers=self._WORKERS, max_q_size=self._MAX_Q_SIZE)
caption_results = []
datum_results = []
for _ in tqdm(range(steps_per_epoch)):
generator_output = None
while enqueuer.is_running():
if not enqueuer.queue.empty():
generator_output = enqueuer.queue.get()
break
else:
sleep(self._WAIT_TIME)
X, y, datum_batch = generator_output
captions_pred_str = self._predict_batch(X, y)
caption_results += captions_pred_str
datum_results += datum_batch
enqueuer.stop()
if include_datum:
return zip(caption_results, datum_results)
else:
return caption_results
def _predict(self,
data_generator_function,
steps_per_epoch,
include_datum=True):
data_generator = data_generator_function(include_datum=True)
enqueuer = GeneratorEnqueuer(data_generator, pickle_safe=False)
enqueuer.start(workers=self._WORKERS, max_q_size=self._MAX_Q_SIZE)
caption_results = []
datum_results = []
for _ in tqdm(range(steps_per_epoch)):
generator_output = None
while enqueuer.is_running():
if not enqueuer.queue.empty():
generator_output = enqueuer.queue.get()
break
else:
sleep(self._WAIT_TIME)
X, y, datum_batch = generator_output
captions_pred_str = self._predict_batch(X, y)
caption_results += captions_pred_str
datum_results += datum_batch
enqueuer.stop()
if include_datum:
return zip(caption_results, datum_results)
else:
return caption_results
def get_batch(num_workers=10, **kwargs):
try:
enqueuer = GeneratorEnqueuer(generator(**kwargs), pickle_safe=True)
enqueuer.start(max_q_size=24, workers=num_workers)
generator_output = None
while True:
while enqueuer.is_running():
if not enqueuer.queue.empty():
# print self.enqueuer.queue.qsize()
generator_output = enqueuer.queue.get()
break
else:
time.sleep(0.01)
yield generator_output
generator_output = None
finally:
if enqueuer is not None:
enqueuer.stop()
def get_batch(num_workers=10, **kwargs):
try:
enqueuer = GeneratorEnqueuer(generator(**kwargs), pickle_safe=True)
enqueuer.start(max_q_size=24, workers=num_workers)
generator_output = None
while True:
while enqueuer.is_running():
if not enqueuer.queue.empty():
# print self.enqueuer.queue.qsize()
generator_output = enqueuer.queue.get()
break
else:
time.sleep(0.01)
yield generator_output
generator_output = None
finally:
if enqueuer is not None:
enqueuer.stop()
def predict(self, test_gen, tag='pred'):
if self.cf.pred_model:
# TODO model predict method for other tasks
print('\n > Predicting the model...')
if self.cf.problem_type == 'segmentation':
# Load best trained model
self.model.load_weights(self.cf.weights_file)
# Create output directory
if not os.path.exists(
os.path.join(self.cf.savepath, 'Predictions')):
os.makedirs(os.path.join(self.cf.savepath, 'Predictions'))
# Create a data generator
enqueuer = GeneratorEnqueuer(test_gen, wait_time=0.05)
enqueuer.start(workers=1, max_queue_size=1)
output_generator = enqueuer.get()
# Process the dataset
start_time = time.time()
for i in range(
int(
math.ceil(self.cf.dataset.n_images_test /
float(self.cf.batch_size_test)))):
# Get data for this minibatch
data = next(output_generator)
x_true = data[0]
y_true = data[1].astype('int32')
# Get prediction for this minibatch
y_pred = self.model.predict(x_true)
# Reshape y_true and compute the y_pred argmax
if K.image_dim_ordering() == 'th':
print(
'Predict method not implemented for th dim ordering.'
)
return
else:
# Find the most probable class of each pixel
y_pred = np.argmax(y_pred, axis=2)
y_true = np.argmax(y_true, axis=2)
# Reshape from (?,172800) to (?, 360, 480)
# print('Acc with shapes y_pred={} and y_true{}: {}'.format(y_pred.shape,y_true.shape,np.mean(np.ravel(y_pred==y_true))))
y_true = np.reshape(y_true,
(x_true.shape[0], x_true.shape[1],
x_true.shape[2]))
y_pred = np.reshape(y_pred,
(x_true.shape[0], x_true.shape[1],
x_true.shape[2]))
# print('Acc with shapes y_pred={} and y_true{}: {}'.format(y_pred.shape,y_true.shape,np.mean(np.ravel(y_pred==y_true))))
# Save output images
save_img3(x_true, y_true, y_pred,
os.path.join(self.cf.savepath, 'Predictions'), 0,
self.cf.dataset.color_map,
self.cf.dataset.classes, tag + str(i),
self.cf.dataset.void_class)
# Stop data generator
if enqueuer is not None:
enqueuer.stop()
else:
print('Predict method not implemented.')
return
total_time = time.time() - start_time
fps = float(self.cf.dataset.n_images_test) / total_time
s_p_f = total_time / float(self.cf.dataset.n_images_test)
print(' Predicting time: {}. FPS: {}. Seconds per Frame: {}'.
format(total_time, fps, s_p_f))
def fit_generator(self, generator,
steps_per_epoch,
epochs=1,
verbose=1,
callbacks=None,
validation_data=None,
validation_steps=None,
class_weight=None,
max_q_size=10,
workers=1,
pickle_safe=False,
initial_epoch=0):
"""Fits the model on data yielded batch-by-batch by a Python generator.
The generator is run in parallel to the model, for efficiency.
For instance, this allows you to do real-time data augmentation
on images on CPU in parallel to training your model on GPU.
# Arguments
generator: a generator.
The output of the generator must be either
- a tuple (inputs, targets)
- a tuple (inputs, targets, sample_weights).
All arrays should contain the same number of samples.
The generator is expected to loop over its data
indefinitely. An epoch finishes when `steps_per_epoch`
samples have been seen by the model.
steps_per_epoch: Total number of steps (batches of samples)
to yield from `generator` before declaring one epoch
finished and starting the next epoch. It should typically
be equal to the number of unique samples if your dataset
divided by the batch size.
epochs: integer, total number of iterations on the data.
verbose: verbosity mode, 0, 1, or 2.
callbacks: list of callbacks to be called during training.
validation_data: this can be either
- a generator for the validation data
- a tuple (inputs, targets)
- a tuple (inputs, targets, sample_weights).
validation_steps: Only relevant if `validation_data`
is a generator. Total number of steps (batches of samples)
to yield from `generator` before stopping.
class_weight: dictionary mapping class indices to a weight
for the class.
max_q_size: maximum size for the generator queue
workers: maximum number of processes to spin up
when using process based threading
pickle_safe: if True, use process based threading.
Note that because
this implementation relies on multiprocessing,
you should not pass
non picklable arguments to the generator
as they can't be passed
easily to children processes.
initial_epoch: epoch at which to start training
(useful for resuming a previous training run)
# Returns
A `History` object.
# Example
```python
def generate_arrays_from_file(path):
while 1:
f = open(path)
for line in f:
# create numpy arrays of input data
# and labels, from each line in the file
x1, x2, y = process_line(line)
yield ({'input_1': x1, 'input_2': x2}, {'output': y})
f.close()
model.fit_generator(generate_arrays_from_file('/my_file.txt'),
steps_per_epoch=10000, epochs=10)
```
# Raises
ValueError: In case the generator yields
data in an invalid format.
"""
wait_time = 0.01 # in seconds
epoch = initial_epoch
do_validation = bool(validation_data)
self._make_train_function()
if do_validation:
self._make_test_function()
# python 2 has 'next', 3 has '__next__'
# avoid any explicit version checks
val_gen = (hasattr(validation_data, 'next') or
hasattr(validation_data, '__next__'))
if val_gen and not validation_steps:
raise ValueError('When using a generator for validation data, '
'you must specify a value for '
'`validation_steps`.')
out_labels = self.metrics_names
callback_metrics = out_labels + ['val_' + n for n in out_labels]
# prepare callbacks
self.history = cbks.History()
callbacks = [cbks.BaseLogger()] + (callbacks or []) + [self.history]
if verbose:
callbacks += [cbks.ProgbarLogger(count_mode='steps')]
callbacks = cbks.CallbackList(callbacks)
# it's possible to callback a different model than self:
if hasattr(self, 'callback_model') and self.callback_model:
callback_model = self.callback_model
else:
callback_model = self
callbacks.set_model(callback_model)
callbacks.set_params({
'epochs': epochs,
'steps': steps_per_epoch,
'verbose': verbose,
'do_validation': do_validation,
'metrics': callback_metrics,
})
callbacks.on_train_begin()
if do_validation and not val_gen:
if len(validation_data) == 2:
val_x, val_y = validation_data
val_sample_weight = None
elif len(validation_data) == 3:
val_x, val_y, val_sample_weight = validation_data
else:
raise ValueError('validation_data should be a tuple '
'`(val_x, val_y, val_sample_weight)` '
'or `(val_x, val_y)`. Found: ' +
str(validation_data))
val_x, val_y, val_sample_weights = self._standardize_user_data(
val_x, val_y, val_sample_weight)
for cbk in callbacks:
cbk.validation_data = val_x + [val_y, val_sample_weights]
enqueuer = None
try:
enqueuer = GeneratorEnqueuer(generator, pickle_safe=pickle_safe)
enqueuer.start(max_q_size=max_q_size, workers=workers)
callback_model.stop_training = False
while epoch < epochs:
callbacks.on_epoch_begin(epoch)
steps_done = 0
batch_index = 0
while steps_done < steps_per_epoch:
generator_output = None
while enqueuer.is_running() or not enqueuer.queue.empty():
if not enqueuer.queue.empty():
generator_output = enqueuer.queue.get()
break
else:
time.sleep(wait_time)
# Added by ZZ.
if not enqueuer.is_running() and enqueuer.queue.empty() and generator_output is None:
continue
if not hasattr(generator_output, '__len__'):
raise ValueError('output of generator should be '
'a tuple `(x, y, sample_weight)` '
'or `(x, y)`. Found: ' +
str(generator_output))
if len(generator_output) == 2:
x, y = generator_output
sample_weight = None
elif len(generator_output) == 3:
x, y, sample_weight = generator_output
else:
raise ValueError('output of generator should be '
'a tuple `(x, y, sample_weight)` '
'or `(x, y)`. Found: ' +
str(generator_output))
# build batch logs
batch_logs = {}
if isinstance(x, list):
batch_size = x[0].shape[0]
elif isinstance(x, dict):
batch_size = list(x.values())[0].shape[0]
else:
batch_size = x.shape[0]
batch_logs['batch'] = batch_index
batch_logs['size'] = batch_size
callbacks.on_batch_begin(batch_index, batch_logs)
outs = self.train_on_batch(x, y,
sample_weight=sample_weight,
class_weight=class_weight)
if not isinstance(outs, list):
outs = [outs]
for l, o in zip(out_labels, outs):
batch_logs[l] = o
callbacks.on_batch_end(batch_index, batch_logs)
# Construct epoch logs.
epoch_logs = {}
batch_index += 1
steps_done += 1
# Epoch finished.
if steps_done >= steps_per_epoch and do_validation:
if val_gen:
val_outs = self.evaluate_generator(
validation_data,
validation_steps,
max_q_size=max_q_size,
workers=workers,
pickle_safe=pickle_safe)
else:
# No need for try/except because
# data has already been validated.
val_outs = self.evaluate(
val_x, val_y,
batch_size=batch_size,
sample_weight=val_sample_weights,
verbose=0)
if not isinstance(val_outs, list):
val_outs = [val_outs]
# Same labels assumed.
for l, o in zip(out_labels, val_outs):
epoch_logs['val_' + l] = o
callbacks.on_epoch_end(epoch, epoch_logs)
epoch += 1
if callback_model.stop_training:
break
finally:
if enqueuer is not None:
enqueuer.stop()
callbacks.on_train_end()
return self.history
def predict(self, test_gen, tag='pred', prob=False, weights=None):
if self.cf.pred_model:
print('Predict method not implemented.')
return
# TODO fix model predict method
print('\n > Predicting the model...')
# Load best trained model
# self.model.load_weights(os.path.join(self.cf.savepath, "weights.hdf5"))
if weights == None:
self.model.load_weights(self.cf.weights_file)
else:
self.model.load_weights(weights)
# Create a data generator
enqueuer = GeneratorEnqueuer(test_gen)
enqueuer.start(max_q_size=1, nb_worker=1)
# data_gen_queue, _stop, _generator_threads = GeneratorEnqueuer(test_gen, max_q_size=1)
# Process the dataset
start_time = time.time()
predictions = np.zeros(
(self.cf.dataset.n_images_test, self.cf.target_size_test[0],
self.cf.target_size_test[1], self.cf.dataset.n_classes))
true = np.zeros(
(self.cf.dataset.n_images_test, self.cf.target_size_test[0],
self.cf.target_size_test[1], 1))
# for _ in range(int(math.ceil(self.cf.dataset.n_images_train/float(self.cf.batch_size_test)))):
for i in range(
int(
math.ceil(self.cf.dataset.n_images_test /
float(self.cf.batch_size_test)))):
# Get data for this minibatch
# data = enqueuer.queue.get()
while enqueuer.is_running():
if not enqueuer.queue.empty():
generator_output = enqueuer.queue.get()
break
else:
time.sleep(0.01)
x_true = generator_output[0]
y_true = generator_output[1].astype('int32')
# Get prediction for this minibatch
y_pred = self.model.predict(x_true)
# Compute the argmax
if not prob:
y_pred = np.argmax(y_pred, axis=1)
predictions[i:i + y_pred.shape[0]] = y_pred
true[i:i + y_pred.shape[0]] = y_true
# # Reshape y_true
# y_true = np.reshape(y_true, (y_true.shape[0], y_true.shape[2],
# y_true.shape[3]))
#
# save_img3(x_true, y_true, y_pred, self.cf.savepath, 0,
# self.cf.dataset.color_map, self.cf.dataset.classes, tag+str(_), self.cf.dataset.void_class)
# Stop data generator
if enqueuer is not None:
enqueuer.stop()
total_time = time.time() - start_time
fps = float(self.cf.dataset.n_images_test) / total_time
s_p_f = total_time / float(self.cf.dataset.n_images_test)
print(' Predicting time: {}. FPS: {}. Seconds per Frame: {}'.
format(total_time, fps, s_p_f))
return predictions, true, self.model.metrics_names
def on_epoch_end(self, epoch, logs={}):
self.nepoch = epoch
self.losses.append(logs.get('loss'))
self.val_losses.append(logs.get('val_loss'))
enqueuer = GeneratorEnqueuer(self.val_gen, pickle_safe=False)
enqueuer.start(nb_worker=1, max_q_size=5, wait_time=0.05)
plt.ioff()
plt.plot(self.losses, 'b-')
plt.plot(self.val_losses, 'c-')
plt.title('model loss')
plt.ylabel('loss')
plt.xlabel('epoch')
plt.legend(['train', 'test'], loc='upper left')
plt.ylim(0, 0.2)
#plt.draw()
plt.pause(0.02)
plt.savefig(self.directory + '_loss.png')
#print self.losses[0]
#if self.count >100:
# self.count = 0
#pprint(self.model, indent=2)
# Get data for this minibatch
data = None
while enqueuer.is_running():
if not enqueuer.queue.empty():
data = enqueuer.queue.get()
break
else:
plt.pause(0.05)
#data = data_gen_queue.get()
x_true = data[0]
#print (x_true.shape)
img = x_true[0].transpose(1, 2, 0)
# plt.imshow(img)
# plt.show()
# plt.pause(0.01)
# y_true = data[1]
# img = y_true[0].transpose(1,2,0)
# plt.imshow(img)
# plt.show()
# plt.pause(0.01)
#print (y_true.shape)
# Get prediction for this minibatch
y_pred = self.model.predict(x_true)
# Reshape y_true and compute the y_pred argmax
# if K.image_dim_ordering() == 'th':
# y_pred = np.argmax(y_pred, axis=1)
# y_true = np.array(y_true).transpose(0,3,1,2)
# else:
# y_pred = np.argmax(y_pred, axis=3)
for i in range(y_pred.shape[0]):
img = (y_pred[i] - np.min(y_pred[i]))
img = (img / np.max(img)).transpose(1, 2, 0)
#print (img.shape)
#y_pred = y_pred[0]
gt = np.array(x_true[i]).transpose(1, 2, 0)
#print (gt.shape)
#y_pred = y_pred.transpose(1,2,0)
#y_gt = y_gt.transpose(1,2,0)
result = np.concatenate((gt, img), axis=1)
#print np.max(y_pred), np.min(y_pred), np.max(y_gt), np.min(y_gt)
#plt.imshow(result)
#plt.show()
#pause(0.01)
oname = os.path.join(self.directory,
str(self.nepoch) + '_' + str(i) + '.jpg')
#print oname
#time.sleep(5)
#self.count += 1
cv2.imwrite(oname, cv2.cvtColor(result * 255, cv2.COLOR_RGB2BGR))
# y_true = np.reshape(y_true, (y_true.shape[0], y_true.shape[1],
# y_true.shape[2]))
# Save output images
# Stop data generator
if enqueuer is not None:
enqueuer.stop()
'''
val_samples =20
pred = self.model.predict_generator(self.val_gen, val_samples)
#plt.imshow(y_pred[0,:,:,:].transpose(1,2,0))
#plt.show()
#plt.pause(0.01)
#print (y_pred.shape)
#y_pred = self.model.predict(self.model.validation_data[0][self.count:self.count+1,:,:,:])
#print y_pred.shape
#print type(y_pred)
for i in range(val_samples):
y_pred = (pred[i,:,:,:] - np.min(pred[i,:,:,:]))
y_pred = y_pred / np.max(y_pred)
y_pred = y_pred[0]
y_gt = model.validation_data[1][self.count]
y_pred = y_pred.transpose(1,2,0)
y_gt = y_gt.transpose(1,2,0)
result = np.concatenate((y_gt,y_pred ), axis=1)
#print np.max(y_pred), np.min(y_pred), np.max(y_gt), np.min(y_gt)
#plt.imshow(result)
#plt.show()
#pause(0.01)
oname = os.path.join(self.directory, str(self.count)+'_'+str(self.nepoch)+'_'+'.jpg')
#print oname
#time.sleep(5)
self.count += 1
cv2.imwrite(oname, result*255)
'''
return
for _ in range(5):
x_test, y_test = my_gen_train.next()
y_pred = netG.predict(x_test)
draw_batch_images(x_test, y_pred, index_pro=111, is_save=False)
x_pred = netF.predict(y_test)
draw_batch_images(x_pred, y_test, index_pro=222, is_save=False)
exit()
print("load total samples:" + str(data_total_len))
nb_batches = int(data_total_len / batch_size)
progress_bar = Progbar(target=nb_batches)
try:
enqueuer = GeneratorEnqueuer(my_gen_train, pickle_safe=pickle_safe)
enqueuer.start(max_q_size=max_q_size, workers=workers)
true_dis_sample = np.ones((batch_size, patch_size, patch_size, 1))
false_dis_sample = np.zeros(
(batch_size, patch_size, patch_size, 1))
true_false_dis_sample = np.concatenate(
[true_dis_sample, false_dis_sample])
for epoch in range(nb_epochs):
print('Epoch {} of {}'.format(epoch, nb_epochs))
print(" ")
for index in range(0, nb_batches):
progress_bar.update(index)
generator_output = None
while enqueuer.is_running():
def predict(self, test_gen, tag='pred', max_q_size=10, workers=1, pickle_safe=False, wait_time = 0.01):
if self.cf.pred_model and test_gen is not None:
print('\n > Predicting the model...')
aux = 'image_result'
result_path = os.path.join(self.cf.savepath,aux)
if not os.path.exists(result_path):
os.makedirs(result_path)
# Load best trained model
# self.model.load_weights(os.path.join(self.cf.savepath, "weights.hdf5"))
self.model.load_weights(self.cf.weights_file)
if self.cf.problem_type == 'detection':
priors = self.cf.dataset.priors
anchors = np.array(priors)
thresh = 0.6
nms_thresh = 0.3
classes = self.cf.dataset.classes
# Create a data generator
data_gen_queue = GeneratorEnqueuer(test_gen, pickle_safe=pickle_safe)
data_gen_queue.start(workers, max_q_size)
# Process the dataset
start_time = time.time()
image_counter = 1
for _ in range(int(math.ceil(self.cf.dataset.n_images_train/float(self.cf.batch_size_test)))):
data = None
while data_gen_queue.is_running():
if not data_gen_queue.queue.empty():
data = data_gen_queue.queue.get()
break
else:
time.sleep(wait_time)
x_true = data[0]
y_true = data[1].astype('int32')
# Get prediction for this minibatch
y_pred = self.model.predict(x_true)
if self.cf.model_name == "yolo" or self.cf.model_name == "tiny-yolo" or self.cf.model_name == "yolt":
for i in range(len(y_pred)):
#Process the YOLO output to obtain final BBox per image
boxes = yolo_postprocess_net_out(y_pred[i], anchors, classes, thresh, nms_thresh)
#Draw the Bbox in the image to visualize
im = yolo_draw_detections(boxes, x_true[i], anchors, classes, thresh, nms_thresh)
out_name = os.path.join(result_path, 'img_' + str(image_counter).zfill(4)+ '.png')
scipy.misc.toimage(im).save(out_name)
image_counter = image_counter+1
elif self.cf.model_name == "ssd":
results = self.cf.bbox_util.detection_out(y_pred)
for j in range(len(results)):
# Parse the outputs.
if np.any(results[j]):
det_label = results[j][:, 0]
det_conf = results[j][:, 1]
det_xmin = results[j][:, 2]
det_ymin = results[j][:, 3]
det_xmax = results[j][:, 4]
det_ymax = results[j][:, 5]
# Get detections with confidence higher than 0.6.
top_indices = [i for i, conf in enumerate(det_conf) if conf >= thresh]
top_conf = det_conf[top_indices]
top_label_indices = det_label[top_indices].tolist()
top_xmin = det_xmin[top_indices]
top_ymin = det_ymin[top_indices]
top_xmax = det_xmax[top_indices]
top_ymax = det_ymax[top_indices]
out_name = os.path.join(result_path, 'img_' + str(image_counter).zfill(4)+ '.png')
if top_indices:
im = self.cf.bbox_util.ssd_draw_detections(top_conf, top_label_indices, top_xmin, top_ymin,
top_xmax, top_ymax, x_true[j], classes, out_name)
#scipy.misc.toimage(im).save(out_name)
#im.savefig(out_name)
image_counter = image_counter+1
else:
raise ValueError("No model name defined or valid: " + self.model_name)
# Stop data generator
data_gen_queue.stop()
total_time = time.time() - start_time
fps = float(self.cf.dataset.n_images_test) / total_time
s_p_f = total_time / float(self.cf.dataset.n_images_test)
print (' Predicting time: {}. FPS: {}. Seconds per Frame: {}'.format(total_time, fps, s_p_f))