2 # svrt is the ``Synthetic Visual Reasoning Test'', an image
3 # generator for evaluating classification performance of machine
4 # learning systems, humans and primates.
6 # Copyright (c) 2017 Idiap Research Institute, http://www.idiap.ch/
7 # Written by Francois Fleuret <francois.fleuret@idiap.ch>
9 # This file is part of svrt.
11 # svrt is free software: you can redistribute it and/or modify it
12 # under the terms of the GNU General Public License version 3 as
13 # published by the Free Software Foundation.
15 # svrt is distributed in the hope that it will be useful, but
16 # WITHOUT ANY WARRANTY; without even the implied warranty of
17 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 # General Public License for more details.
20 # You should have received a copy of the GNU General Public License
21 # along with selector. If not, see <http://www.gnu.org/licenses/>.
25 from multiprocessing import Pool, cpu_count
27 from torch import Tensor
28 from torch.autograd import Variable
32 ######################################################################
34 def generate_one_batch(s):
36 target = torch.LongTensor(self.batch_size).bernoulli_(0.5)
37 input = svrt.generate_vignettes(problem_number, target)
38 input = input.float().view(input.size(0), 1, input.size(1), input.size(2))
41 target = target.cuda()
42 return [ input, target ]
46 def __init__(self, problem_number, nb_batches, batch_size, cuda = False):
48 self.batch_size = batch_size
49 self.problem_number = problem_number
50 self.nb_batches = nb_batches
51 self.nb_samples = self.nb_batches * self.batch_size
53 seed_list = torch.LongTensor(self.nb_batches).random_().tolist()
56 # for b in range(0, self.nb_batches):
57 # self.data.append(generate_one_batch(seed_list[b]))
59 self.data = Pool(cpu_count()).map(generate_one_batch, seed_list)
63 for b in range(0, self.nb_batches):
64 input = self.data[b][0]
65 acc += input.sum() / input.numel()
66 acc_sq += input.pow(2).sum() / input.numel()
68 mean = acc / self.nb_batches
69 std = sqrt(acc_sq / self.nb_batches - mean * mean)
70 for b in range(0, self.nb_batches):
71 self.data[b][0].sub_(mean).div_(std)
73 def get_batch(self, b):
76 ######################################################################
78 class CompressedVignetteSet:
79 def __init__(self, problem_number, nb_batches, batch_size, cuda = False):
81 self.batch_size = batch_size
82 self.problem_number = problem_number
83 self.nb_batches = nb_batches
84 self.nb_samples = self.nb_batches * self.batch_size
86 self.input_storages = []
90 for b in range(0, self.nb_batches):
91 target = torch.LongTensor(self.batch_size).bernoulli_(0.5)
92 input = svrt.generate_vignettes(problem_number, target)
93 acc += input.float().sum() / input.numel()
94 acc_sq += input.float().pow(2).sum() / input.numel()
95 self.targets.append(target)
96 self.input_storages.append(svrt.compress(input.storage()))
98 self.mean = acc / self.nb_batches
99 self.std = sqrt(acc_sq / self.nb_batches - self.mean * self.mean)
101 def get_batch(self, b):
102 input = torch.ByteTensor(svrt.uncompress(self.input_storages[b])).float()
103 input = input.view(self.batch_size, 1, 128, 128).sub_(self.mean).div_(self.std)
104 target = self.targets[b]
108 target = target.cuda()
112 ######################################################################