def run_model(name, learning_rate, batch_size, epoch):
#make model
sess = tf.Session()
md = models.Model(sess, name, FLAGS)
#data preprocessing
dp = utils.preprocess(FLAGS)
x_data, y_data, x_test, y_test = dp.Mnist2data()
#initialize
sess.run(tf.global_variables_initializer())
#training
for ep in range(epoch):
avg_cost = 0
total_batches = int(x_data.shape[0]/batch_size)
#print(int(x_data.shape[0]), total_batches, batch_size)
for i in range(total_batches):
x,y = x_data[i*batch_size:i*batch_size+batch_size], y_data[i*batch_size:i*batch_size+batch_size]
cost, _ = md.train(x,y)
avg_cost += cost/batch_size
print('[*] epoch:', '%04d' % (ep + 1), ', cost =', '{:.9f}'.format(avg_cost))
print("[*] learning Finishied!")
#testset predict
print('Accuracy:', md.get_accuracy(x_test, y_test))
return True
def _load_from_disk(self, image_files, states, outputs):
embnet_images, embnet_states, embnet_outputs = [], [], []
for i in range(self.support_query_size):
images, emb_states, emb_outputs = self.data_sequencer.load(
image_files[i], states[i], outputs[i])
# images will be of shape (sequence, w, h, 3)
embnet_images.append(images)
embnet_states.append(emb_states)
embnet_outputs.append(emb_outputs)
embed_images = tf.stack(embnet_images)
embnet_states = tf.stack(embnet_states)
embnet_outputs = tf.stack(embnet_outputs)
embnet_states.set_shape(
(self.support_query_size, self.data_sequencer.frames, None))
embnet_outputs.set_shape(
(self.support_query_size, self.data_sequencer.frames, None))
# Grab a random timestep in one of the support and query trajectories
ctrnet_timestep = tf.random_uniform(
(2, ), 0, self.dataset.time_horizon, tf.int32)
# The first should be a support and the last should be a query
ctrnet_images = [
utils.tf_load_image(image_files[0], ctrnet_timestep[0]),
utils.tf_load_image(image_files[-1], ctrnet_timestep[1])
]
ctrnet_states = [
states[0][ctrnet_timestep[0]], states[-1][ctrnet_timestep[1]]
]
ctrnet_outputs = [
outputs[0][ctrnet_timestep[0]], outputs[-1][ctrnet_timestep[1]]
]
ctrnet_images = tf.stack(ctrnet_images)
ctrnet_states = tf.stack(ctrnet_states)
ctrnet_outputs = tf.stack(ctrnet_outputs)
embed_images = utils.preprocess(embed_images)
ctrnet_images = utils.preprocess(ctrnet_images)
return (embed_images, embnet_states, embnet_outputs, ctrnet_images,
ctrnet_states, ctrnet_outputs)
def predict(self, imgs, sizes=None, visualize=False):
# this function gives the prediction results of an image input
if visualize:
self.use_preset('visualize')
prepared_imgs = list()
sizes = list()
for img in imgs:
size = img.shape[1:]
img = preprocess(tonumpy(img))
prepared_imgs.append(img)
sizes.append(size)
else:
prepared_imgs = imgs
bboxes = list()
labels = list()
scores = list()
for img, size in zip(prepared_imgs, sizes):
img = totensor(img[None]).float()
scale = img.shape[3] / size[1]
roi_cls_loc, roi_scores, rois, _ = self.forward(img, scale=scale)
roi_score = roi_scores.data
roi_cls_loc = roi_cls_loc.data
roi = totensor(rois) / scale
mean = t.Tensor(self.loc_normalize_mean).cuda(). \
repeat(self.n_class)[None]
std = t.Tensor(self.loc_normalize_std).cuda(). \
repeat(self.n_class)[None]
roi_cls_loc = (roi_cls_loc * std + mean)
roi_cls_loc = roi_cls_loc.view(-1, self.n_class, 4)
roi = roi.view(-1, 1, 4).expand_as(roi_cls_loc)
cls_bbox = loc2bbox(
tonumpy(roi).reshape((-1, 4)),
tonumpy(roi_cls_loc).reshape((-1, 4)))
cls_bbox = totensor(cls_bbox)
cls_bbox = cls_bbox.view(-1, self.n_class * 4)
# clip bounding box
cls_bbox[:, 0::2] = (cls_bbox[:, 0::2]).clamp(min=0, max=size[0])
cls_bbox[:, 1::2] = (cls_bbox[:, 1::2]).clamp(min=0, max=size[1])
prob = (F.softmax(totensor(roi_score), dim=1))
bbox, label, score = self._suppress(cls_bbox, prob)
bboxes.append(bbox)
labels.append(label)
scores.append(score)
self.use_preset('evaluate')
return bboxes, labels, scores
def predict(self, imgs, sizes=None, visualize=False):
"""Detect objects from images.
This method predicts objects for each image.
Args:
imgs (iterable of numpy.ndarray): Arrays holding images.
All images are in CHW and RGB format
and the range of their value is :math:`[0, 255]`.
sizes (iterable of ints) input image sizes.
visualize: if use visusalize to show images result.
Returns:
tuple of lists:
This method returns a tuple of three lists,
:obj:`(bboxes, labels, scores)`.
* **bboxes**: A list of float arrays of shape :math:`(R, 4)`, \
where :math:`R` is the number of bounding boxes in a image. \
Each bouding box is organized by \
:math:`(y_{min}, x_{min}, y_{max}, x_{max})` \
in the second axis.
* **labels** : A list of integer arrays of shape :math:`(R,)`. \
Each value indicates the class of the bounding box. \
Values are in range :math:`[0, L - 1]`, where :math:`L` is the \
number of the foreground classes.
* **scores** : A list of float arrays of shape :math:`(R,)`. \
Each value indicates how confident the prediction is.
"""
self.eval()
# Visualize a few results getting from origin images.
if visualize:
self.use_present('visualize')
prepared_imgs = list()
sizes = list()
for img in imgs:
size = img.shape[1:]
img = preprocess(img)
img = torch.from_numpy(img)
prepared_imgs.append(img)
sizes.append(size)
else:
prepared_imgs = imgs
bboxes = list()
labels = list()
scores = list()
for img, size in zip(prepared_imgs, sizes):
img = Variable(img[None].cuda(), volatile=True)
scale = img.shape[3] / size[1]
roi_cls_loc, roi_scores, rois, _ = self.forward(img, scale=scale)
# We are assuming that batch size is 1
roi_score = roi_scores.data
roi_cls_loc = roi_cls_loc.data
roi = torch.from_numpy(rois).cuda() / scale
# Convert predictions to bbox in image coordinates.
# Bounding boxes are scaled to the scale of the input images.
mean = torch.FloatTensor(self.loc_normalize_mean).cuda().repeat(self.n_class)[None]
std = torch.FloatTensor(self.loc_normalize_std).cuda().repeat(self.n_class)[None]
roi_cls_loc = (roi_cls_loc * std + mean)
roi_cls_loc = roi_cls_loc.view(-1, self.n_class, 4)
roi = roi.view(-1, 1, 4).expand_as(roi_cls_loc)
cls_bbox = loc2bbox(roi.cpu().numpy().reshape((-1, 4)),
roi_cls_loc.cpu().numpy().reshape((-1, 4)))
cls_bbox = torch.from_numpy(cls_bbox).view(-1, self.n_class * 4)
# Clip bbox.
cls_bbox[:, 0::2] = (cls_bbox[:, 0::2]).clamp(min=0, max=size[0])
cls_bbox[:, 1::2] = (cls_bbox[:, 1::2]).clamp(min=0, max=size[1])
raw_cls_bbox = cls_bbox.numpy()
raw_prob = F.softmax(Variable(roi_score), dim=1).cpu().data.numpy()
bbox, label, score = self._supress(raw_cls_bbox, raw_prob)
bboxes.append(bbox)
labels.append(label)
scores.append(score)
self.use_present('evaluate')
return bboxes, labels, scores
def consume(self, inputs):
if self.disk_images:
# (Examples,)
input_image = tf.placeholder(tf.string, (self.support, ))
else:
# (Examples, timesteps)
input_image = tf.placeholder(tf.float32,
(None, None) + self.dataset.img_shape)
input_states = tf.placeholder(
tf.float32,
(self.support, self.dataset.time_horizon, self.dataset.state_size))
input_outputs = tf.placeholder(
tf.float32, (self.support, self.dataset.time_horizon,
self.dataset.action_size))
# (B. W, H, C)
input_ctr_image = tf.placeholder(tf.float32,
(None, 1) + self.dataset.img_shape)
input_ctr_state = tf.placeholder(tf.float32,
(None, 1, self.dataset.state_size))
training = tf.placeholder_with_default(False, None)
stacked_embnet_images, bs, cs = [], [], []
for i in range(self.support):
embnet_images, embnet_states, embnet_outputs = (
self.data_sequencer.load(input_image[i], input_states[i],
input_outputs[i]))
embnet_images = utils.preprocess(embnet_images)
stacked_embnet_images.append(embnet_images)
bs.append(embnet_states)
cs.append(embnet_outputs)
embnet_images = tf.stack(stacked_embnet_images)
embnet_images = tf.expand_dims(embnet_images,
axis=0) # set batchsize 1
embnet_states = tf.stack(bs)
embnet_states = tf.expand_dims(embnet_states, axis=0)
embnet_outputs = tf.stack(cs)
embnet_outputs = tf.expand_dims(embnet_outputs, axis=0)
embnet_images.set_shape((None, None, self.data_sequencer.frames) +
self.dataset.img_shape)
embnet_states.set_shape(
(None, None, self.data_sequencer.frames, self.dataset.state_size))
embnet_outputs.set_shape(
(None, None, self.data_sequencer.frames, self.dataset.action_size))
return {
'embnet_images': embnet_images,
'embnet_states': embnet_states,
'embnet_outputs': embnet_outputs,
'input_image_files': input_image,
'input_states': input_states,
'input_outputs': input_outputs,
'ctrnet_images': input_ctr_image,
'ctrnet_states': input_ctr_state,
'training': training,
'support': tf.placeholder_with_default(self.support, None),
'query': tf.placeholder_with_default(0, None),
}
def preproc_dataset(self):
"""Preprocesses dataset according to specified format (see utils)"""
if (self.ds_name == 'wn18' or self.ds_name == 'fb15'):
data_utils.preprocess(self.path)
import logging
import time
from config import CONFIG
from data.utils import preprocess
from gensim.models import KeyedVectors
import gensim.downloader as api
_LOGGER = logging.getLogger(__name__)
_LOGGER.setLevel(logging.INFO)
if __name__ == '__main__':
WV_MODEL = api.load('glove-wiki-gigaword-300')
_LOGGER.info("Starting data preprocessing")
start = time.time()
preprocess("{}/data/all.txt".format(CONFIG['base_dir']), "{}/data/cleaned.txt".format(CONFIG['base_dir']), WV_MODEL.vocab)
end = time.time()
_LOGGER.info("Took {:.2f} seconds to preprocess file".format(end-start))