Skip to content
Permalink
Browse files

Add a version for readPngImageAndPreprocess that can also take a tens…

…or as input. (#2896)

Summary:
*Description*:
Implement a version for readPngImageAndPreprocess that can also take a tensor as input.
*Testing*:
Add a test case in ImageTest: readPngImageAndPreprocessWithAndWithoutInputTensor() that verifies the new version works correctly.
[Optional Fixes #issue]
#2358
Pull Request resolved: #2896

Differential Revision: D15343817

Pulled By: ZchiPitt

fbshipit-source-id: c010e08190fe4625bc9d4484dd1b89c1437cf5c3
  • Loading branch information...
ZchiPitt authored and facebook-github-bot committed May 16, 2019
1 parent 71f2ec2 commit 818cba22d2efff5ebb302aa4f333fdecee965f5c
Showing with 67 additions and 2 deletions.
  1. +20 −1 include/glow/Base/Image.h
  2. +13 −1 lib/Base/Image.cpp
  3. +34 −0 tests/unittests/ImageTest.cpp
@@ -93,7 +93,8 @@ bool writePngImage(Tensor *T, const char *filename,
llvm::ArrayRef<float> mean = zeroMean,
llvm::ArrayRef<float> stddev = oneStd);

/// Read a png image and preprocess it according to several parameters.
/// Read a png image and preprocess it according to several parameters. Create a
/// tensor and store the preprocessed image data into this tensor.
/// \param filename the png file to read.
/// \param imageNormMode normalize values to this range.
/// \param imageChannelOrder the order of color channels.
@@ -107,6 +108,24 @@ Tensor readPngImageAndPreprocess(llvm::StringRef filename,
llvm::ArrayRef<float> mean = zeroMean,
llvm::ArrayRef<float> stddev = oneStd);

/// Read a png image and preprocess it according to several parameters. Take a
/// tensor as a parameter and store the preprocessed image data into this
/// tensor.
/// \param imageData the tensor into which the preprocessed image data
/// will be stored.
/// \param filename the png file to read.
/// \param imageNormMode normalize values to this range.
/// \param imageChannelOrder the order of color channels.
/// \param imageLayout the order of dimensions (channel, height, and width).
/// \param mean use special mean to normalize.
/// \param stdev use special stddev to normalize.
void readPngImageAndPreprocess(Tensor &imageData, llvm::StringRef filename,
ImageNormalizationMode imageNormMode,
ImageChannelOrder imageChannelOrder,
ImageLayout imageLayout,
llvm::ArrayRef<float> mean = zeroMean,
llvm::ArrayRef<float> stddev = oneStd);

/// Loads and normalizes all PNGs into a tensor in the NHWC format with the
/// requested channel ordering.
/// \param filenames list of filenames to read.
@@ -362,6 +362,19 @@ Tensor glow::readPngImageAndPreprocess(llvm::StringRef filename,
llvm::ArrayRef<float> mean,
llvm::ArrayRef<float> stddev) {
Tensor imageData;
readPngImageAndPreprocess(imageData, filename, imageNormMode,
imageChannelOrder, imageLayout, mean, stddev);
return imageData;
}

void glow::readPngImageAndPreprocess(Tensor &imageData,
llvm::StringRef filename,
ImageNormalizationMode imageNormMode,
ImageChannelOrder imageChannelOrder,
ImageLayout imageLayout,
llvm::ArrayRef<float> mean,
llvm::ArrayRef<float> stddev) {

auto range = normModeToRange(imageNormMode);
bool loadSuccess =
!readPngImage(&imageData, filename.data(), range, mean, stddev);
@@ -391,7 +404,6 @@ Tensor glow::readPngImageAndPreprocess(llvm::StringRef filename,
imageData.transpose(&transposed, {2u, 0u, 1u});
imageData = std::move(transposed);
}
return imageData;
}

void glow::loadImagesAndPreprocess(const llvm::ArrayRef<std::string> &filenames,
@@ -55,6 +55,40 @@ TEST(Image, readBadImages) {
ASSERT_FALSE(loadSuccess);
}

TEST(Image, readPngImageAndPreprocessWithAndWithoutInputTensor) {
auto image1 = readPngImageAndPreprocess(
"tests/images/imagenet/cat_285.png", ImageNormalizationMode::k0to1,
ImageChannelOrder::RGB, ImageLayout::NHWC, imagenetNormMean,
imagenetNormStd);
Tensor image2;
readPngImageAndPreprocess(image2, "tests/images/imagenet/cat_285.png",
ImageNormalizationMode::k0to1,
ImageChannelOrder::BGR, ImageLayout::NCHW,
imagenetNormMean, imagenetNormStd);

// Test if the preprocess actually happened.
size_t imgHeight = image1.dims()[0];
size_t imgWidth = image1.dims()[1];
size_t numChannels = image1.dims()[2];

Tensor transposed;
image2.transpose(&transposed, {1u, 2u, 0u});
image2 = std::move(transposed);

Tensor swizzled(image1.getType());
auto IH = image1.getHandle();
auto SH = swizzled.getHandle();
for (size_t z = 0; z < numChannels; z++) {
for (size_t y = 0; y < imgHeight; y++) {
for (size_t x = 0; x < imgWidth; x++) {
SH.at({x, y, numChannels - 1 - z}) = IH.at({x, y, z});
}
}
}
image1 = std::move(swizzled);
EXPECT_TRUE(image1.isEqual(image2, 0.01));
}

TEST(Image, writePngImage) {
auto range = std::make_pair(0.f, 1.f);
Tensor localCopy;

0 comments on commit 818cba2

Please sign in to comment.
You can’t perform that action at this time.