Skip to content
This repository has been archived by the owner on Nov 17, 2023. It is now read-only.

Commit

Permalink
Add seed_aug parameter for ImageRecordItr to fix random seed for defa…
Browse files Browse the repository at this point in the history 
…ult augmentation (#11247)

* add seed_aug parameter for ImageRecordItr to fix random seed for augmentation

* remove white space

* add test

* fix test

* improve according to review: using dmlc::optional and has_value()

* missing header

* change data type and way to get value
  • Loading branch information
@wenyangchu @zhreshold
wenyangchu authored and zhreshold committed Jun 20, 2018
1 parent ba9784d commit 9ffc03c
Show file tree
Hide file tree
Showing 2 changed files with 148 additions and 31 deletions.
12 changes: 12 additions & 0 deletions src/io/image_aug_default.cc
View file
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,7 @@
* \brief Default augmenter.
*/
#include <mxnet/base.h>
#include <dmlc/optional.h>
#include <utility>
#include <string>
#include <algorithm>
Expand Down Expand Up @@ -96,6 +97,9 @@ struct DefaultImageAugmentParam : public dmlc::Parameter<DefaultImageAugmentPara
int pad;
/*! \brief shape of the image data*/
TShape data_shape;
/*! \brief random seed for augmentations */
dmlc::optional<int> seed_aug;

// declare parameters
DMLC_DECLARE_PARAMETER(DefaultImageAugmentParam) {
DMLC_DECLARE_FIELD(resize).set_default(-1)
Expand Down Expand Up @@ -184,6 +188,8 @@ struct DefaultImageAugmentParam : public dmlc::Parameter<DefaultImageAugmentPara
DMLC_DECLARE_FIELD(pad).set_default(0)
.describe("Change size from ``[width, height]`` into "
"``[pad + width + pad, pad + height + pad]`` by padding pixes");
DMLC_DECLARE_FIELD(seed_aug).set_default(dmlc::optional<int>())
.describe("Random seed for augmentations.");
}
};

Expand All @@ -204,6 +210,7 @@ class DefaultImageAugmenter : public ImageAugmenter {
// contructor
DefaultImageAugmenter() {
rotateM_ = cv::Mat(2, 3, CV_32F);
seed_init_state = false;
}
void Init(const std::vector<std::pair<std::string, std::string> >& kwargs) override {
std::vector<std::pair<std::string, std::string> > kwargs_left;
Expand Down Expand Up @@ -244,6 +251,10 @@ class DefaultImageAugmenter : public ImageAugmenter {
}
cv::Mat Process(const cv::Mat &src, std::vector<float> *label,
common::RANDOM_ENGINE *prnd) override {
if (!seed_init_state && param_.seed_aug.has_value()) {
prnd->seed(param_.seed_aug.value());
seed_init_state = true;
}
using mshadow::index_t;
bool is_cropped = false;

Expand Down Expand Up @@ -550,6 +561,7 @@ class DefaultImageAugmenter : public ImageAugmenter {
DefaultImageAugmentParam param_;
/*! \brief list of possible rotate angle */
std::vector<int> rotate_list_;
bool seed_init_state;
};

ImageAugmenter* ImageAugmenter::Create(const std::string& name) {
Expand Down
167 changes: 136 additions & 31 deletions tests/python/unittest/test_io.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -20,7 +20,8 @@
from mxnet.test_utils import *
from mxnet.base import MXNetError
import numpy as np
import os, gzip
import os
import gzip
import pickle as pickle
import time
try:
Expand All @@ -31,16 +32,17 @@
from common import assertRaises
import unittest


def test_MNISTIter():
# prepare data
get_mnist_ubyte()

batch_size = 100
train_dataiter = mx.io.MNISTIter(
image="data/train-images-idx3-ubyte",
label="data/train-labels-idx1-ubyte",
data_shape=(784,),
batch_size=batch_size, shuffle=1, flat=1, silent=0, seed=10)
image="data/train-images-idx3-ubyte",
label="data/train-labels-idx1-ubyte",
data_shape=(784,),
batch_size=batch_size, shuffle=1, flat=1, silent=0, seed=10)
# test_loop
nbatch = 60000 / batch_size
batch_count = 0
Expand All @@ -60,18 +62,19 @@ def test_MNISTIter():
label_1 = train_dataiter.getlabel().asnumpy().flatten()
assert(sum(label_0 - label_1) == 0)


def test_Cifar10Rec():
get_cifar10()
dataiter = mx.io.ImageRecordIter(
path_imgrec="data/cifar/train.rec",
mean_img="data/cifar/cifar10_mean.bin",
rand_crop=False,
and_mirror=False,
shuffle=False,
data_shape=(3,28,28),
batch_size=100,
preprocess_threads=4,
prefetch_buffer=1)
path_imgrec="data/cifar/train.rec",
mean_img="data/cifar/cifar10_mean.bin",
rand_crop=False,
and_mirror=False,
shuffle=False,
data_shape=(3, 28, 28),
batch_size=100,
preprocess_threads=4,
prefetch_buffer=1)
labelcount = [0 for i in range(10)]
batchcount = 0
for batch in dataiter:
Expand All @@ -84,23 +87,26 @@ def test_Cifar10Rec():
for i in range(10):
assert(labelcount[i] == 5000)


def test_NDArrayIter():
data = np.ones([1000, 2, 2])
label = np.ones([1000, 1])
for i in range(1000):
data[i] = i / 100
label[i] = i / 100
dataiter = mx.io.NDArrayIter(data, label, 128, True, last_batch_handle='pad')
dataiter = mx.io.NDArrayIter(
data, label, 128, True, last_batch_handle='pad')
batchidx = 0
for batch in dataiter:
batchidx += 1
assert(batchidx == 8)
dataiter = mx.io.NDArrayIter(data, label, 128, False, last_batch_handle='pad')
dataiter = mx.io.NDArrayIter(
data, label, 128, False, last_batch_handle='pad')
batchidx = 0
labelcount = [0 for i in range(10)]
for batch in dataiter:
label = batch.label[0].asnumpy().flatten()
assert((batch.data[0].asnumpy()[:,0,0] == label).all())
assert((batch.data[0].asnumpy()[:, 0, 0] == label).all())
for i in range(label.shape[0]):
labelcount[int(label[i])] += 1

Expand All @@ -110,6 +116,7 @@ def test_NDArrayIter():
else:
assert(labelcount[i] == 100)


def test_NDArrayIter_h5py():
if not h5py:
return
Expand All @@ -128,17 +135,19 @@ def test_NDArrayIter_h5py():
f.create_dataset("data", data=data)
f.create_dataset("label", data=label)

dataiter = mx.io.NDArrayIter(f["data"], f["label"], 128, True, last_batch_handle='pad')
dataiter = mx.io.NDArrayIter(
f["data"], f["label"], 128, True, last_batch_handle='pad')
batchidx = 0
for batch in dataiter:
batchidx += 1
assert(batchidx == 8)

dataiter = mx.io.NDArrayIter(f["data"], f["label"], 128, False, last_batch_handle='pad')
dataiter = mx.io.NDArrayIter(
f["data"], f["label"], 128, False, last_batch_handle='pad')
labelcount = [0 for i in range(10)]
for batch in dataiter:
label = batch.label[0].asnumpy().flatten()
assert((batch.data[0].asnumpy()[:,0,0] == label).all())
assert((batch.data[0].asnumpy()[:, 0, 0] == label).all())
for i in range(label.shape[0]):
labelcount[int(label[i])] += 1

Expand All @@ -153,6 +162,7 @@ def test_NDArrayIter_h5py():
else:
assert(labelcount[i] == 100)


def test_NDArrayIter_csr():
# creating toy data
num_rows = rnd.randint(5, 15)
Expand All @@ -163,25 +173,28 @@ def test_NDArrayIter_csr():
dns = csr.asnumpy()

# CSRNDArray or scipy.sparse.csr_matrix with last_batch_handle not equal to 'discard' will throw NotImplementedError
assertRaises(NotImplementedError, mx.io.NDArrayIter, {'data': csr}, dns, batch_size)
assertRaises(NotImplementedError, mx.io.NDArrayIter,
{'data': csr}, dns, batch_size)
try:
import scipy.sparse as spsp
train_data = spsp.csr_matrix(dns)
assertRaises(NotImplementedError, mx.io.NDArrayIter, {'data': train_data}, dns, batch_size)
assertRaises(NotImplementedError, mx.io.NDArrayIter,
{'data': train_data}, dns, batch_size)
except ImportError:
pass

# CSRNDArray with shuffle
csr_iter = iter(mx.io.NDArrayIter({'csr_data': csr, 'dns_data': dns}, dns, batch_size,
shuffle=True, last_batch_handle='discard'))
shuffle=True, last_batch_handle='discard'))
num_batch = 0
for batch in csr_iter:
num_batch += 1

assert(num_batch == num_rows // batch_size)

# make iterators
csr_iter = iter(mx.io.NDArrayIter(csr, csr, batch_size, last_batch_handle='discard'))
csr_iter = iter(mx.io.NDArrayIter(
csr, csr, batch_size, last_batch_handle='discard'))
begin = 0
for batch in csr_iter:
expected = np.zeros((batch_size, num_cols))
Expand All @@ -192,6 +205,7 @@ def test_NDArrayIter_csr():
assert_almost_equal(batch.data[0].asnumpy(), expected)
begin += batch_size


def test_LibSVMIter():

def check_libSVMIter_synthetic():
Expand All @@ -214,8 +228,8 @@ def check_libSVMIter_synthetic():
data_train = mx.io.LibSVMIter(data_libsvm=data_path, label_libsvm=label_path,
data_shape=(3, ), label_shape=(3, ), batch_size=3)

first = mx.nd.array([[ 0.5, 0., 1.2], [ 0., 0., 0.], [ 0.6, 2.4, 1.2]])
second = mx.nd.array([[ 0., 0., -1.2], [ 0.5, 0., 1.2], [ 0., 0., 0.]])
first = mx.nd.array([[0.5, 0., 1.2], [0., 0., 0.], [0.6, 2.4, 1.2]])
second = mx.nd.array([[0., 0., -1.2], [0.5, 0., 1.2], [0., 0., 0.]])
i = 0
for batch in iter(data_train):
expected = first.asnumpy() if i == 0 else second.asnumpy()
Expand Down Expand Up @@ -286,10 +300,13 @@ def test_DataBatch():
from nose.tools import ok_
from mxnet.io import DataBatch
import re
batch = DataBatch(data=[mx.nd.ones((2,3))])
ok_(re.match('DataBatch: data shapes: \[\(2L?, 3L?\)\] label shapes: None', str(batch)))
batch = DataBatch(data=[mx.nd.ones((2,3)), mx.nd.ones((7,8))], label=[mx.nd.ones((4,5))])
ok_(re.match('DataBatch: data shapes: \[\(2L?, 3L?\), \(7L?, 8L?\)\] label shapes: \[\(4L?, 5L?\)\]', str(batch)))
batch = DataBatch(data=[mx.nd.ones((2, 3))])
ok_(re.match(
'DataBatch: data shapes: \[\(2L?, 3L?\)\] label shapes: None', str(batch)))
batch = DataBatch(data=[mx.nd.ones((2, 3)), mx.nd.ones(
(7, 8))], label=[mx.nd.ones((4, 5))])
ok_(re.match(
'DataBatch: data shapes: \[\(2L?, 3L?\), \(7L?, 8L?\)\] label shapes: \[\(4L?, 5L?\)\]', str(batch)))


def test_CSVIter():
Expand All @@ -307,7 +324,7 @@ def check_CSVIter_synthetic(dtype='float32'):
for i in range(1000):
fout.write('0\n')

data_train = mx.io.CSVIter(data_csv=data_path, data_shape=(8,8),
data_train = mx.io.CSVIter(data_csv=data_path, data_shape=(8, 8),
label_csv=label_path, batch_size=100, dtype=dtype)
expected = mx.nd.ones((100, 8, 8), dtype=dtype) * int(entry_str)
for batch in iter(data_train):
Expand All @@ -318,6 +335,93 @@ def check_CSVIter_synthetic(dtype='float32'):
for dtype in ['int32', 'float32']:
check_CSVIter_synthetic(dtype=dtype)


def test_ImageRecordIter_seed_augmentation():
get_cifar10()
seed_aug = 3

# check whether to get constant images after fixing seed_aug
dataiter = mx.io.ImageRecordIter(
path_imgrec="data/cifar/train.rec",
mean_img="data/cifar/cifar10_mean.bin",
shuffle=False,
data_shape=(3, 28, 28),
batch_size=3,
rand_crop=True,
rand_mirror=True,
max_random_scale=1.3,
max_random_illumination=3,
max_rotate_angle=10,
random_l=50,
random_s=40,
random_h=10,
max_shear_ratio=2,
seed_aug=seed_aug)
batch = dataiter.next()
data = batch.data[0].asnumpy().astype(np.uint8)

dataiter = mx.io.ImageRecordIter(
path_imgrec="data/cifar/train.rec",
mean_img="data/cifar/cifar10_mean.bin",
shuffle=False,
data_shape=(3, 28, 28),
batch_size=3,
rand_crop=True,
rand_mirror=True,
max_random_scale=1.3,
max_random_illumination=3,
max_rotate_angle=10,
random_l=50,
random_s=40,
random_h=10,
max_shear_ratio=2,
seed_aug=seed_aug)
batch = dataiter.next()
data2 = batch.data[0].asnumpy().astype(np.uint8)
assert(np.array_equal(data,data2))

# check whether to get different images after change seed_aug
dataiter = mx.io.ImageRecordIter(
path_imgrec="data/cifar/train.rec",
mean_img="data/cifar/cifar10_mean.bin",
shuffle=False,
data_shape=(3, 28, 28),
batch_size=3,
rand_crop=True,
rand_mirror=True,
max_random_scale=1.3,
max_random_illumination=3,
max_rotate_angle=10,
random_l=50,
random_s=40,
random_h=10,
max_shear_ratio=2,
seed_aug=seed_aug+1)
batch = dataiter.next()
data2 = batch.data[0].asnumpy().astype(np.uint8)
assert(not np.array_equal(data,data2))

# check whether seed_aug changes the iterator behavior
dataiter = mx.io.ImageRecordIter(
path_imgrec="data/cifar/train.rec",
mean_img="data/cifar/cifar10_mean.bin",
shuffle=False,
data_shape=(3, 28, 28),
batch_size=3)
batch = dataiter.next()
data = batch.data[0].asnumpy().astype(np.uint8)

dataiter = mx.io.ImageRecordIter(
path_imgrec="data/cifar/train.rec",
mean_img="data/cifar/cifar10_mean.bin",
shuffle=False,
data_shape=(3, 28, 28),
batch_size=3,
seed_aug=seed_aug)
batch = dataiter.next()
data2 = batch.data[0].asnumpy().astype(np.uint8)
assert(np.array_equal(data,data2))

if __name__ == "__main__":
test_NDArrayIter()
if h5py:
Expand All @@ -327,3 +431,4 @@ def check_CSVIter_synthetic(dtype='float32'):
test_LibSVMIter()
test_NDArrayIter_csr()
test_CSVIter()
test_ImageRecordIter_seed_augmentation()

0 comments on commit 9ffc03c

Please sign in to comment.