297 lines
15 KiB
Python
Executable file
297 lines
15 KiB
Python
Executable file
from data import *
|
|
import numpy.random as nr
|
|
import numpy as n
|
|
import random as r
|
|
from time import time
|
|
from threading import Thread
|
|
from math import sqrt
|
|
import sys
|
|
from pylab import *
|
|
from PIL import Image
|
|
from StringIO import StringIO
|
|
|
|
class JPEGBatchLoaderThread(Thread):
|
|
def __init__(self, data_dir, path, freq_to_id, tgt, tgt_labels, list_out):
|
|
Thread.__init__(self)
|
|
self.data_dir = data_dir
|
|
self.path = path
|
|
self.tgt = tgt
|
|
self.tgt_labels = tgt_labels
|
|
self.list_out = list_out
|
|
self.freq_to_id = freq_to_id
|
|
#print "loading %d" % self.bnum
|
|
|
|
@staticmethod
|
|
def raw_to_freq_id(raw_tags, freq_to_id):
|
|
raw_tags = [''.join(t.lower().strip().split()) for t in raw_tags]
|
|
return [freq_to_id[t] for t in raw_tags if t in freq_to_id]
|
|
|
|
@staticmethod
|
|
def load_jpeg_batch((strings, sizes, labels), freq_to_id, tgt, tgt_labels):
|
|
tgt_labels[:] = 0
|
|
for k,s in enumerate(strings):
|
|
ima = n.asarray(Image.open(StringIO(s)).convert('RGB'))
|
|
tgt[k,:] = ima.swapaxes(0,2).swapaxes(1,2).flatten()
|
|
tgt_labels[k, JPEGBatchLoaderThread.raw_to_freq_id(labels[k], freq_to_id)] = 1
|
|
|
|
return {'data': tgt[:len(strings),:],
|
|
'labels': tgt_labels[:len(strings),:]}
|
|
|
|
def run(self):
|
|
p = self.load_jpeg_batch(unpickle(self.path),
|
|
self.freq_to_id,
|
|
self.tgt,
|
|
self.tgt_labels)
|
|
self.list_out.append(p)
|
|
|
|
class ColorNoiseMakerThread(Thread):
|
|
def __init__(self, pca_stdevs, pca_vecs, num_noise, list_out):
|
|
Thread.__init__(self)
|
|
self.pca_stdevs, self.pca_vecs = pca_stdevs, pca_vecs
|
|
self.num_noise = num_noise
|
|
self.list_out = list_out
|
|
|
|
def run(self):
|
|
noise = n.dot(nr.randn(self.num_noise, 3).astype(n.single) * self.pca_stdevs.T, self.pca_vecs.T)
|
|
self.list_out.append(noise)
|
|
|
|
class FlickrDP(LabeledDataProvider):
|
|
MAX_PCA_COMPONENTS = 1024 # Use this many components for noise generation
|
|
def __init__(self, data_dir, batch_range, init_epoch=1, init_batchnum=None, dp_params={}, test=False):
|
|
LabeledDataProvider.__init__(self, data_dir, batch_range, init_epoch, init_batchnum, dp_params, test)
|
|
self.init_commons(data_dir, batch_range, init_epoch, init_batchnum, dp_params, test)
|
|
|
|
def init_commons(self, data_dir, batch_range, init_epoch=1, init_batchnum=None, dp_params={}, test=False):
|
|
self.data_mean = self.batch_meta['data_mean'].astype(n.single)
|
|
self.color_eig = self.batch_meta['color_pca'][1].astype(n.single)
|
|
self.color_stdevs = n.c_[self.batch_meta['color_pca'][0].astype(n.single)]
|
|
self.color_noise_coeff = dp_params['color_noise']
|
|
self.pca_noise_coeff = dp_params['pca_noise']
|
|
self.num_colors = 3
|
|
self.img_size = int(sqrt(self.batch_meta['num_vis'] / self.num_colors))
|
|
self.freq_to_id = self.batch_meta['freq_to_id']
|
|
|
|
def get_labels(self, datadic):
|
|
pass
|
|
|
|
def showimg(self, img):
|
|
pixels = img.shape[0] / 3
|
|
size = int(sqrt(pixels))
|
|
img = img.reshape((3,size,size)).swapaxes(0,2).swapaxes(0,1)
|
|
imshow(img, interpolation='nearest')
|
|
show()
|
|
|
|
def get_next_batch(self):
|
|
epoch, batchnum, datadic = LabeledDataProvider.get_next_batch(self)
|
|
# This takes about 1 sec per batch :(
|
|
# If I don't convert both to single ahead of time, it takes even longer.
|
|
data = n.require(datadic['data'] - self.data_mean, dtype=n.single, requirements='C')
|
|
|
|
labels = self.get_labels(datadic)
|
|
|
|
# Labels have to be in the range 0-(number of classes - 1)
|
|
assert labels.max() < self.get_num_classes(), "Invalid labels!"
|
|
assert labels.min() >= 0, "Invalid labels!"
|
|
return epoch, batchnum, [data, labels]
|
|
|
|
|
|
# Takes as input an array returned by get_next_batch
|
|
# Returns a (numCases, imgSize, imgSize, 3) array which can be
|
|
# fed to pylab for plotting.
|
|
# This is used by shownet.py to plot test case predictions.
|
|
def get_plottable_data(self, data, add_mean=True):
|
|
return n.require((data + (self.data_mean if add_mean else 0)).reshape(data.shape[0], 3, self.img_size, self.img_size).swapaxes(1,3).swapaxes(1,2) / 255.0, dtype=n.single)
|
|
|
|
class JPEGCroppedFlickrDP(FlickrDP):
|
|
def __init__(self, data_dir, batch_range=None, init_epoch=1, init_batchnum=None, dp_params=None, test=False):
|
|
LabeledDataProvider.__init__(self, data_dir, batch_range, init_epoch, init_batchnum, dp_params, test)
|
|
self.init_commons(data_dir, batch_range, init_epoch, init_batchnum, dp_params, test)
|
|
|
|
self.img_size = int(sqrt(self.batch_meta['num_vis'] / self.num_colors))
|
|
self.border_size = dp_params['crop_border']
|
|
self.inner_size = self.img_size - self.border_size*2
|
|
self.multiview = dp_params['multiview_test'] and test
|
|
self.num_views = 5*2
|
|
self.data_mult = self.num_views if self.multiview else 1
|
|
self.crop_chunk = 32 # This many images will be cropped in the same way
|
|
self.batch_size = self.batch_meta['batch_size']
|
|
|
|
# Maintain poitners to previously-returned data matrices so they don't get garbage collected.
|
|
# I've never seen this happen but it's a safety measure.
|
|
self.data = [None, None]
|
|
self.cropped_data = [n.zeros((0*self.data_mult, self.get_data_dims()), dtype=n.float32) for x in xrange(2)]
|
|
if self.test:
|
|
self.orig_data = [n.zeros((self.batch_size, self.img_size**2*3), dtype=n.uint8) for x in xrange(1)]
|
|
self.orig_labels = [n.zeros((self.batch_size, self.get_num_classes()), dtype=n.float32) for x in xrange(2)]
|
|
else:
|
|
self.orig_data = [n.zeros((self.batch_size, self.img_size**2*3), dtype=n.uint8) for x in xrange(2)]
|
|
# There have to be 3 copies of labels because this matrix actually gets used by the training code
|
|
self.orig_labels = [n.zeros((self.batch_size, self.get_num_classes()), dtype=n.float32) for x in xrange(3)]
|
|
|
|
self.loader_thread, self.color_noise_thread = None, None
|
|
self.convnet = dp_params['convnet']
|
|
|
|
self.num_noise = self.batch_size
|
|
self.batches_generated, self.loaders_started = 0, 0
|
|
self.data_mean_crop = self.data_mean.reshape((3,self.img_size,self.img_size))[:,self.border_size:self.border_size+self.inner_size,self.border_size:self.border_size+self.inner_size].reshape((1,3*self.inner_size**2))
|
|
|
|
def get_data_dims(self, idx=0):
|
|
assert idx in (0,1), "Invalid index: %d" % idx
|
|
if idx == 0:
|
|
return self.inner_size**2 * 3
|
|
return self.get_num_classes()
|
|
|
|
def start_loader(self, batch_idx):
|
|
self.load_data = []
|
|
#print "loading %d" % self.batch_range_perm[self.batch_idx]
|
|
self.loader_thread = JPEGBatchLoaderThread(self.data_dir, self.get_data_file_name(self.batch_range[batch_idx]), self.freq_to_id,
|
|
self.orig_data[self.loaders_started % 2], self.orig_labels[self.loaders_started % 3],
|
|
self.load_data)
|
|
self.loader_thread.start()
|
|
self.loaders_started += 1
|
|
|
|
def start_color_noise_maker(self):
|
|
color_noise_list = []
|
|
self.color_noise_thread = ColorNoiseMakerThread(self.color_stdevs, self.color_eig, self.num_noise, color_noise_list)
|
|
self.color_noise_thread.start()
|
|
return color_noise_list
|
|
|
|
def get_labels(self, datadic):
|
|
pass
|
|
|
|
def get_next_batch(self):
|
|
self.d_idx = self.batches_generated % 2
|
|
if self.test:
|
|
epoch, batchnum, self.data[self.d_idx] = LabeledDataProvider.get_next_batch(self)
|
|
self.data[self.d_idx] = JPEGBatchLoaderThread.load_jpeg_batch(self.data[self.d_idx], self.freq_to_id, self.orig_data[0], self.orig_labels[self.d_idx])
|
|
else:
|
|
epoch, batchnum = self.curr_epoch, self.curr_batchnum
|
|
|
|
if self.loader_thread is None:
|
|
self.start_loader(self.batch_idx)
|
|
self.loader_thread.join()
|
|
self.data[self.d_idx] = self.load_data[0]
|
|
|
|
self.start_loader(self.get_next_batch_idx())
|
|
else:
|
|
# Set the argument to join to 0 to re-enable batch reuse
|
|
self.loader_thread.join()
|
|
if not self.loader_thread.is_alive():
|
|
self.data[self.d_idx] = self.load_data[0]
|
|
self.start_loader(self.get_next_batch_idx())
|
|
# else:
|
|
# print "Re-using batch"
|
|
self.advance_batch()
|
|
|
|
cropped = self.get_cropped_data(self.data[self.d_idx])
|
|
if self.color_noise_coeff > 0 and not self.test:
|
|
# At this point the data already has 0 mean.
|
|
# So I'm going to add noise to it, but I'm also going to scale down
|
|
# the original data. This is so that the overall scale of the training
|
|
# data doesn't become too different from the test data.
|
|
s = cropped.shape
|
|
cropped_size = self.get_data_dims(0) / 3
|
|
ncases = s[0]
|
|
|
|
if self.color_noise_thread is None:
|
|
self.color_noise_list = self.start_color_noise_maker()
|
|
self.color_noise_thread.join()
|
|
self.color_noise = self.color_noise_list[0]
|
|
self.color_noise_list = self.start_color_noise_maker()
|
|
else:
|
|
self.color_noise_thread.join(0)
|
|
if not self.color_noise_thread.is_alive():
|
|
self.color_noise = self.color_noise_list[0]
|
|
self.color_noise_list = self.start_color_noise_maker()
|
|
|
|
cropped = self.cropped_data[self.d_idx] = cropped.reshape((ncases*3, cropped_size))
|
|
self.color_noise = self.color_noise[:ncases,:].reshape((3*ncases, 1))
|
|
cropped += self.color_noise * self.color_noise_coeff
|
|
cropped = self.cropped_data[self.d_idx] = cropped.reshape((ncases, 3* cropped_size))
|
|
cropped /= (1.0 + self.color_noise_coeff)
|
|
|
|
self.data[self.d_idx]['labels'] = self.get_labels(self.data[self.d_idx])
|
|
self.data[self.d_idx]['data'] = cropped
|
|
self.batches_generated += 1
|
|
|
|
# idx = 1000
|
|
# cropped -= cropped.min()
|
|
# cropped /= cropped.max()
|
|
#
|
|
# print [self.batch_meta['label_names'][i] for i in n.where(self.data['labels'][idx,:]==1)[0]]
|
|
# self.showimg(cropped[idx,:])
|
|
#print cropped.shape
|
|
return epoch, batchnum, [self.data[self.d_idx]['data'].T, self.data[self.d_idx]['labels'].T]
|
|
|
|
def get_cropped_data(self, data):
|
|
cropped = self.cropped_data[self.d_idx]
|
|
if cropped.shape[0] != data['data'].shape[0] * self.data_mult:
|
|
cropped = self.cropped_data[self.d_idx] = n.zeros((data['data'].shape[0] * self.data_mult, cropped.shape[1]), dtype=n.float32)
|
|
self.__trim_borders(data['data'], cropped)
|
|
|
|
return self.subtract_mean(cropped)
|
|
|
|
def subtract_mean(self,data):
|
|
data -= self.data_mean_crop
|
|
return data
|
|
|
|
# Takes as input an array returned by get_next_batch
|
|
# Returns a (numCases, imgSize, imgSize, 3) array which can be
|
|
# fed to pylab for plotting.
|
|
# This is used by shownet.py to plot test case predictions.
|
|
def get_plottable_data(self, data, add_mean=True):
|
|
return n.require((data.T + (self.data_mean_crop if add_mean else 0)).reshape(data.shape[1], 3, self.inner_size, self.inner_size).swapaxes(1,3).swapaxes(1,2) / 255.0, dtype=n.single)
|
|
|
|
def __trim_borders(self, x, target):
|
|
y = x.reshape(x.shape[0], 3, self.img_size, self.img_size)
|
|
|
|
if self.test: # don't need to loop over cases
|
|
if self.multiview:
|
|
start_positions = [(0,0), (0, self.border_size*2),
|
|
(self.border_size, self.border_size),
|
|
(self.border_size*2, 0), (self.border_size*2, self.border_size*2)]
|
|
end_positions = [(sy+self.inner_size, sx+self.inner_size) for (sy,sx) in start_positions]
|
|
for i in xrange(self.num_views/2):
|
|
pic = y[:,:,start_positions[i][0]:end_positions[i][0],start_positions[i][1]:end_positions[i][1]]
|
|
target[i * x.shape[0]:(i+1)* x.shape[0],:] = pic.reshape((x.shape[0], self.get_data_dims()))
|
|
target[(self.num_views/2 + i) * x.shape[0]:(self.num_views/2 +i+1)* x.shape[0],:] = pic[:,:,:,::-1].reshape((x.shape[0],self.get_data_dims()))
|
|
else:
|
|
pic = y[:,:,self.border_size:self.border_size+self.inner_size,self.border_size:self.border_size+self.inner_size] # just take the center for now
|
|
target[:,:] = pic.reshape((x.shape[0], self.get_data_dims()))
|
|
else:
|
|
for c in xrange(0, x.shape[0], self.crop_chunk): # loop over cases in chunks
|
|
startY, startX = nr.randint(0,self.border_size*2 + 1), nr.randint(0,self.border_size*2 + 1)
|
|
|
|
endY, endX = startY + self.inner_size, startX + self.inner_size
|
|
c_end = min(c + self.crop_chunk, x.shape[0])
|
|
pic = y[c:c_end,:,startY:endY,startX:endX]
|
|
if nr.randint(2) == 0: # also flip the images with 50% probability
|
|
pic = pic[:,:,:,::-1]
|
|
target[c:c_end,:] = pic.reshape((c_end-c, self.get_data_dims()))
|
|
#target[:] = n.require(target[:,nr.permutation(x.shape[1])], requirements='C')
|
|
|
|
class JPEGCroppedFlickrCEDP(JPEGCroppedFlickrDP):
|
|
def __init__(self, data_dir, batch_range, init_epoch=1, init_batchnum=None, dp_params={}, test=False):
|
|
JPEGCroppedFlickrDP.__init__(self, data_dir, batch_range, init_epoch, init_batchnum, dp_params, test)
|
|
|
|
def get_labels(self, data):
|
|
return n.require(n.tile(data['labels'], (self.data_mult, 1)), requirements='C')
|
|
|
|
class DummyConvNetCEDP(LabeledDummyDataProvider):
|
|
def __init__(self, data_dim):
|
|
LabeledDummyDataProvider.__init__(self, data_dim, num_classes=16, num_cases=16)
|
|
|
|
def get_next_batch(self):
|
|
epoch, batchnum, dic = LabeledDummyDataProvider.get_next_batch(self)
|
|
|
|
dic['data'] = n.require(dic['data'].T, requirements='F')
|
|
dic['labels'] = n.zeros((self.get_data_dims(idx=1), dic['data'].shape[1]), dtype=n.float32, order='F')
|
|
for c in xrange(dic['labels'].shape[1]): # loop over cases
|
|
r = nr.randint(0, dic['labels'].shape[0])
|
|
dic['labels'][r,c] = 1
|
|
|
|
return epoch, batchnum, [dic['data'], dic['labels']]
|
|
|
|
# Returns the dimensionality of the two data matrices returned by get_next_batch
|
|
def get_data_dims(self, idx=0):
|
|
return self.batch_meta['num_vis'] if idx == 0 else 16
|