pgjmblvch
/
mindspore2022
forked from huawei/mindspore2022
0
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[MD] Transform api decouple

This commit is contained in:
luoyang 2020-11-02 21:23:23 +08:00
parent 1d9f04e096
commit 9a57cbacd9
4 changed files with 989 additions and 141 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

475
mindspore/ccsrc/minddata/dataset/api/vision.cc
View File

@ -32,6 +32,7 @@
#endif
#include "minddata/dataset/kernels/image/decode_op.h"
#ifndef ENABLE_ANDROID
#include "minddata/dataset/kernels/image/equalize_op.h"
#include "minddata/dataset/kernels/image/hwc_to_chw_op.h"
#include "minddata/dataset/kernels/image/invert_op.h"
#include "minddata/dataset/kernels/image/mixup_batch_op.h"
@ -49,7 +50,10 @@
#include "minddata/dataset/kernels/image/random_horizontal_flip_op.h"
#include "minddata/dataset/kernels/image/random_horizontal_flip_with_bbox_op.h"
#include "minddata/dataset/kernels/image/random_posterize_op.h"
#include "minddata/dataset/kernels/image/random_resize_op.h"
#include "minddata/dataset/kernels/image/random_resize_with_bbox_op.h"
#include "minddata/dataset/kernels/image/random_rotation_op.h"
#include "minddata/dataset/kernels/image/random_select_subpolicy_op.h"
#include "minddata/dataset/kernels/image/random_sharpness_op.h"
#include "minddata/dataset/kernels/image/random_solarize_op.h"
#include "minddata/dataset/kernels/image/random_vertical_flip_op.h"
@ -61,6 +65,8 @@
#include "minddata/dataset/kernels/image/resize_with_bbox_op.h"
#include "minddata/dataset/kernels/image/rgba_to_bgr_op.h"
#include "minddata/dataset/kernels/image/rgba_to_rgb_op.h"
#include "minddata/dataset/kernels/image/soft_dvpp/soft_dvpp_decode_random_crop_resize_jpeg_op.h"
#include "minddata/dataset/kernels/image/soft_dvpp/soft_dvpp_decode_resize_jpeg_op.h"
#include "minddata/dataset/kernels/image/swap_red_blue_op.h"
#include "minddata/dataset/kernels/image/uniform_aug_op.h"
#endif
@ -78,10 +84,7 @@ namespace vision {
std::shared_ptr<AutoContrastOperation> AutoContrast(float cutoff, std::vector<uint32_t> ignore) {
auto op = std::make_shared<AutoContrastOperation>(cutoff, ignore);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create BoundingBoxAugmentOperation.
@ -89,20 +92,14 @@ std::shared_ptr<BoundingBoxAugmentOperation> BoundingBoxAugment(std::shared_ptr<
float ratio) {
auto op = std::make_shared<BoundingBoxAugmentOperation>(transform, ratio);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create CenterCropOperation.
std::shared_ptr<CenterCropOperation> CenterCrop(std::vector<int32_t> size) {
auto op = std::make_shared<CenterCropOperation>(size);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
#endif
@ -110,80 +107,63 @@ std::shared_ptr<CenterCropOperation> CenterCrop(std::vector<int32_t> size) {
std::shared_ptr<CropOperation> Crop(std::vector<int32_t> coordinates, std::vector<int32_t> size) {
auto op = std::make_shared<CropOperation>(coordinates, size);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
#ifndef ENABLE_ANDROID
// Function to create CutMixBatchOperation.
std::shared_ptr<CutMixBatchOperation> CutMixBatch(ImageBatchFormat image_batch_format, float alpha, float prob) {
auto op = std::make_shared<CutMixBatchOperation>(image_batch_format, alpha, prob);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create CutOutOp.
std::shared_ptr<CutOutOperation> CutOut(int32_t length, int32_t num_patches) {
auto op = std::make_shared<CutOutOperation>(length, num_patches);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create DecodeOperation.
std::shared_ptr<DecodeOperation> Decode(bool rgb) {
auto op = std::make_shared<DecodeOperation>(rgb);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create EqualizeOperation.
std::shared_ptr<EqualizeOperation> Equalize() {
auto op = std::make_shared<EqualizeOperation>();
// Input validation
return op->ValidateParams() ? op : nullptr;
}
// Function to create HwcToChwOperation.
std::shared_ptr<HwcToChwOperation> HWC2CHW() {
auto op = std::make_shared<HwcToChwOperation>();
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create InvertOperation.
std::shared_ptr<InvertOperation> Invert() {
auto op = std::make_shared<InvertOperation>();
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create MixUpBatchOperation.
std::shared_ptr<MixUpBatchOperation> MixUpBatch(float alpha) {
auto op = std::make_shared<MixUpBatchOperation>(alpha);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create NormalizeOperation.
std::shared_ptr<NormalizeOperation> Normalize(std::vector<float> mean, std::vector<float> std) {
auto op = std::make_shared<NormalizeOperation>(mean, std);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create PadOperation.
@ -191,10 +171,7 @@ std::shared_ptr<PadOperation> Pad(std::vector<int32_t> padding, std::vector<uint
BorderType padding_mode) {
auto op = std::make_shared<PadOperation>(padding, fill_value, padding_mode);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create RandomAffineOperation.
@ -207,20 +184,14 @@ std::shared_ptr<RandomAffineOperation> RandomAffine(const std::vector<float_t> &
auto op = std::make_shared<RandomAffineOperation>(degrees, translate_range, scale_range, shear_ranges, interpolation,
fill_value);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create RandomColorOperation.
std::shared_ptr<RandomColorOperation> RandomColor(float t_lb, float t_ub) {
auto op = std::make_shared<RandomColorOperation>(t_lb, t_ub);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
std::shared_ptr<TensorOp> RandomColorOperation::Build() {
@ -234,10 +205,7 @@ std::shared_ptr<RandomColorAdjustOperation> RandomColorAdjust(std::vector<float>
std::vector<float> saturation, std::vector<float> hue) {
auto op = std::make_shared<RandomColorAdjustOperation>(brightness, contrast, saturation, hue);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create RandomCropOperation.
@ -246,10 +214,7 @@ std::shared_ptr<RandomCropOperation> RandomCrop(std::vector<int32_t> size, std::
BorderType padding_mode) {
auto op = std::make_shared<RandomCropOperation>(size, padding, pad_if_needed, fill_value, padding_mode);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create RandomCropDecodeResizeOperation.
@ -260,10 +225,7 @@ std::shared_ptr<RandomCropDecodeResizeOperation> RandomCropDecodeResize(std::vec
int32_t max_attempts) {
auto op = std::make_shared<RandomCropDecodeResizeOperation>(size, scale, ratio, interpolation, max_attempts);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create RandomCropWithBBoxOperation.
@ -272,40 +234,42 @@ std::shared_ptr<RandomCropWithBBoxOperation> RandomCropWithBBox(std::vector<int3
BorderType padding_mode) {
auto op = std::make_shared<RandomCropWithBBoxOperation>(size, padding, pad_if_needed, fill_value, padding_mode);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create RandomHorizontalFlipOperation.
std::shared_ptr<RandomHorizontalFlipOperation> RandomHorizontalFlip(float prob) {
auto op = std::make_shared<RandomHorizontalFlipOperation>(prob);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create RandomHorizontalFlipOperation.
std::shared_ptr<RandomHorizontalFlipWithBBoxOperation> RandomHorizontalFlipWithBBox(float prob) {
auto op = std::make_shared<RandomHorizontalFlipWithBBoxOperation>(prob);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create RandomPosterizeOperation.
std::shared_ptr<RandomPosterizeOperation> RandomPosterize(const std::vector<uint8_t> &bit_range) {
auto op = std::make_shared<RandomPosterizeOperation>(bit_range);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create RandomResizeOperation.
std::shared_ptr<RandomResizeOperation> RandomResize(std::vector<int32_t> size) {
auto op = std::make_shared<RandomResizeOperation>(size);
// Input validation
return op->ValidateParams() ? op : nullptr;
}
// Function to create RandomResizeWithBBoxOperation.
std::shared_ptr<RandomResizeWithBBoxOperation> RandomResizeWithBBox(std::vector<int32_t> size) {
auto op = std::make_shared<RandomResizeWithBBoxOperation>(size);
// Input validation
return op->ValidateParams() ? op : nullptr;
}
// Function to create RandomResizedCropOperation.
@ -314,10 +278,7 @@ std::shared_ptr<RandomResizedCropOperation> RandomResizedCrop(std::vector<int32_
int32_t max_attempts) {
auto op = std::make_shared<RandomResizedCropOperation>(size, scale, ratio, interpolation, max_attempts);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create RandomRotationOperation.
@ -326,60 +287,50 @@ std::shared_ptr<RandomRotationOperation> RandomRotation(std::vector<float> degre
std::vector<uint8_t> fill_value) {
auto op = std::make_shared<RandomRotationOperation>(degrees, resample, expand, center, fill_value);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create RandomSharpnessOperation.
std::shared_ptr<RandomSharpnessOperation> RandomSharpness(std::vector<float> degrees) {
auto op = std::make_shared<RandomSharpnessOperation>(degrees);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create RandomSolarizeOperation.
std::shared_ptr<RandomSolarizeOperation> RandomSolarize(std::vector<uint8_t> threshold) {
auto op = std::make_shared<RandomSolarizeOperation>(threshold);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create RandomSelectSubpolicyOperation.
std::shared_ptr<RandomSelectSubpolicyOperation> RandomSelectSubpolicy(
std::vector<std::vector<std::pair<std::shared_ptr<TensorOperation>, double>>> policy) {
auto op = std::make_shared<RandomSelectSubpolicyOperation>(policy);
// Input validation
return op->ValidateParams() ? op : nullptr;
}
// Function to create RandomVerticalFlipOperation.
std::shared_ptr<RandomVerticalFlipOperation> RandomVerticalFlip(float prob) {
auto op = std::make_shared<RandomVerticalFlipOperation>(prob);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create RandomVerticalFlipWithBBoxOperation.
std::shared_ptr<RandomVerticalFlipWithBBoxOperation> RandomVerticalFlipWithBBox(float prob) {
auto op = std::make_shared<RandomVerticalFlipWithBBoxOperation>(prob);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create RescaleOperation.
std::shared_ptr<RescaleOperation> Rescale(float rescale, float shift) {
auto op = std::make_shared<RescaleOperation>(rescale, shift);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
#endif
@ -387,10 +338,7 @@ std::shared_ptr<RescaleOperation> Rescale(float rescale, float shift) {
std::shared_ptr<ResizeOperation> Resize(std::vector<int32_t> size, InterpolationMode interpolation) {
auto op = std::make_shared<ResizeOperation>(size, interpolation);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
#ifndef ENABLE_ANDROID
@ -398,40 +346,43 @@ std::shared_ptr<ResizeOperation> Resize(std::vector<int32_t> size, Interpolation
std::shared_ptr<ResizeWithBBoxOperation> ResizeWithBBox(std::vector<int32_t> size, InterpolationMode interpolation) {
auto op = std::make_shared<ResizeWithBBoxOperation>(size, interpolation);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create RgbaToBgrOperation.
std::shared_ptr<RgbaToBgrOperation> RGBA2BGR() {
auto op = std::make_shared<RgbaToBgrOperation>();
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create RgbaToRgbOperation.
std::shared_ptr<RgbaToRgbOperation> RGBA2RGB() {
auto op = std::make_shared<RgbaToRgbOperation>();
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create SoftDvppDecodeRandomCropResizeJpegOperation.
std::shared_ptr<SoftDvppDecodeRandomCropResizeJpegOperation> SoftDvppDecodeRandomCropResizeJpeg(
std::vector<int32_t> size, std::vector<float> scale, std::vector<float> ratio, int32_t max_attempts) {
auto op = std::make_shared<SoftDvppDecodeRandomCropResizeJpegOperation>(size, scale, ratio, max_attempts);
// Input validation
return op->ValidateParams() ? op : nullptr;
}
// Function to create SoftDvppDecodeResizeJpegOperation.
std::shared_ptr<SoftDvppDecodeResizeJpegOperation> SoftDvppDecodeResizeJpeg(std::vector<int32_t> size) {
auto op = std::make_shared<SoftDvppDecodeResizeJpegOperation>(size);
// Input validation
return op->ValidateParams() ? op : nullptr;
}
// Function to create SwapRedBlueOperation.
std::shared_ptr<SwapRedBlueOperation> SwapRedBlue() {
auto op = std::make_shared<SwapRedBlueOperation>();
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
// Function to create UniformAugOperation.
@ -439,10 +390,7 @@ std::shared_ptr<UniformAugOperation> UniformAugment(std::vector<std::shared_ptr<
int32_t num_ops) {
auto op = std::make_shared<UniformAugOperation>(transforms, num_ops);
// Input validation
if (!op->ValidateParams()) {
return nullptr;
}
return op;
return op->ValidateParams() ? op : nullptr;
}
#endif
@ -660,6 +608,7 @@ std::shared_ptr<TensorOp> CropOperation::Build() {
height = size_[0];
width = size_[0];
// User has specified crop_width.
if (size_.size() == 2) {
width = size_[1];
}
@ -722,6 +671,11 @@ Status DecodeOperation::ValidateParams() { return Status::OK(); }
std::shared_ptr<TensorOp> DecodeOperation::Build() { return std::make_shared<DecodeOp>(rgb_); }
// EqualizeOperation
Status EqualizeOperation::ValidateParams() { return Status::OK(); }
std::shared_ptr<TensorOp> EqualizeOperation::Build() { return std::make_shared<EqualizeOp>(); }
// HwcToChwOperation
Status HwcToChwOperation::ValidateParams() { return Status::OK(); }
@ -1417,11 +1371,80 @@ std::shared_ptr<TensorOp> RandomPosterizeOperation::Build() {
return tensor_op;
}
// RandomResizeOperation
RandomResizeOperation::RandomResizeOperation(std::vector<int32_t> size) : size_(size) {}
Status RandomResizeOperation::ValidateParams() {
// size
if (size_.size() != 2 && size_.size() != 1) {
std::string err_msg =
"RandomResize: size must be a vector of one or two values, got: " + std::to_string(size_.size());
MS_LOG(ERROR) << err_msg;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
if (size_[0] <= 0 || (size_.size() == 2 && size_[1] <= 0)) {
std::string err_msg = "RandomResize: size must only contain positive integers.";
MS_LOG(ERROR) << "RandomResize: size must only contain positive integers, got: " << size_;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
return Status::OK();
}
std::shared_ptr<TensorOp> RandomResizeOperation::Build() {
// If size is a single value, the smaller edge of the image will be
// resized to this value with the same image aspect ratio.
int32_t height = size_[0];
int32_t width = 0;
// User specified the width value.
if (size_.size() == 2) {
width = size_[1];
}
std::shared_ptr<RandomResizeOp> tensor_op = std::make_shared<RandomResizeOp>(height, width);
return tensor_op;
}
// RandomResizeWithBBoxOperation
RandomResizeWithBBoxOperation::RandomResizeWithBBoxOperation(std::vector<int32_t> size) : size_(size) {}
Status RandomResizeWithBBoxOperation::ValidateParams() {
// size
if (size_.size() != 2 && size_.size() != 1) {
std::string err_msg =
"RandomResizeWithBBox: size must be a vector of one or two values, got: " + std::to_string(size_.size());
MS_LOG(ERROR) << err_msg;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
if (size_[0] <= 0 || (size_.size() == 2 && size_[1] <= 0)) {
std::string err_msg = "RandomResizeWithBBox: size must only contain positive integers.";
MS_LOG(ERROR) << "RandomResizeWithBBox: size must only contain positive integers, got: " << size_;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
return Status::OK();
}
std::shared_ptr<TensorOp> RandomResizeWithBBoxOperation::Build() {
// If size is a single value, the smaller edge of the image will be
// resized to this value with the same image aspect ratio.
int32_t height = size_[0];
int32_t width = 0;
// User specified the width value.
if (size_.size() == 2) {
width = size_[1];
}
std::shared_ptr<RandomResizeWithBBoxOp> tensor_op = std::make_shared<RandomResizeWithBBoxOp>(height, width);
return tensor_op;
}
// RandomResizedCropOperation
RandomResizedCropOperation::RandomResizedCropOperation(std::vector<int32_t> size, std::vector<float> scale,
std::vector<float> ratio, InterpolationMode interpolation,
int32_t max_attempts)
: size_(size), scale_(scale), ratio_(ratio), interpolation_(interpolation), max_attempts_(max_attempts) {}
Status RandomResizedCropOperation::ValidateParams() {
// size
if (size_.size() != 2 && size_.size() != 1) {
@ -1455,8 +1478,9 @@ Status RandomResizedCropOperation::ValidateParams() {
}
// ratio
if (ratio_.size() != 2) {
std::string err_msg = "RandomResizedCrop: ratio must be in the format of (min, max).";
MS_LOG(ERROR) << "RandomResizedCrop: ratio must be in the format of (min, max), but got: " << ratio_;
std::string err_msg =
"RandomResizedCrop: ratio must be a vector of two values, got: " + std::to_string(ratio_.size());
MS_LOG(ERROR) << err_msg;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
if (ratio_[0] < 0 || ratio_[1] < 0) {
@ -1470,12 +1494,23 @@ Status RandomResizedCropOperation::ValidateParams() {
<< ratio_;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
// max_attempts
if (max_attempts_ < 1) {
std::string err_msg =
"RandomResizedCrop: max_attempts must be greater than or equal to 1, got: " + std::to_string(max_attempts_);
MS_LOG(ERROR) << err_msg;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
return Status::OK();
}
std::shared_ptr<TensorOp> RandomResizedCropOperation::Build() {
int32_t height = size_[0], width = size_[0];
if (size_.size() == 2) width = size_[1];
int32_t height = size_[0];
int32_t width = size_[0];
// User specified the width value.
if (size_.size() == 2) {
width = size_[1];
}
std::shared_ptr<RandomCropAndResizeOp> tensor_op = std::make_shared<RandomCropAndResizeOp>(
height, width, scale_[0], scale_[1], ratio_[0], ratio_[1], interpolation_, max_attempts_);
return tensor_op;
@ -1549,6 +1584,55 @@ std::shared_ptr<TensorOp> RandomRotationOperation::Build() {
return tensor_op;
}
// RandomSelectSubpolicyOperation.
RandomSelectSubpolicyOperation::RandomSelectSubpolicyOperation(
std::vector<std::vector<std::pair<std::shared_ptr<TensorOperation>, double>>> policy)
: policy_(policy) {}
Status RandomSelectSubpolicyOperation::ValidateParams() {
if (policy_.empty()) {
std::string err_msg = "RandomSelectSubpolicy: policy must not be empty";
MS_LOG(ERROR) << err_msg;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
for (int32_t i = 0; i < policy_.size(); i++) {
if (policy_[i].empty()) {
std::string err_msg = "RandomSelectSubpolicy: policy[" + std::to_string(i) + "] must not be empty";
MS_LOG(ERROR) << err_msg;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
for (int32_t j = 0; j < policy_[i].size(); j++) {
if (policy_[i][j].first == nullptr) {
std::string transform_pos = "[" + std::to_string(i) + "]" + "[" + std::to_string(j) + "]";
std::string err_msg = "RandomSelectSubpolicy: transform in policy" + transform_pos + " must not be null";
MS_LOG(ERROR) << err_msg;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
if (policy_[i][j].second < 0.0 || policy_[i][j].second > 1.0) {
std::string transform_pos = "[" + std::to_string(i) + "]" + "[" + std::to_string(j) + "]";
std::string err_msg = "RandomSelectSubpolicy: probability of transform in policy" + transform_pos +
" must be between 0.0 and 1.0, got: " + std::to_string(policy_[i][j].second);
MS_LOG(ERROR) << err_msg;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
}
}
return Status::OK();
}
std::shared_ptr<TensorOp> RandomSelectSubpolicyOperation::Build() {
std::vector<Subpolicy> policy_tensor_ops;
for (int32_t i = 0; i < policy_.size(); i++) {
Subpolicy sub_policy_tensor_ops;
for (int32_t j = 0; j < policy_[i].size(); j++) {
sub_policy_tensor_ops.push_back(std::make_pair(policy_[i][j].first->Build(), policy_[i][j].second));
}
policy_tensor_ops.push_back(sub_policy_tensor_ops);
}
std::shared_ptr<RandomSelectSubpolicyOp> tensor_op = std::make_shared<RandomSelectSubpolicyOp>(policy_tensor_ops);
return tensor_op;
}
// Function to create RandomSharpness.
RandomSharpnessOperation::RandomSharpnessOperation(std::vector<float> degrees) : degrees_(degrees) {}
@ -1669,6 +1753,8 @@ Status ResizeOperation::ValidateParams() {
}
std::shared_ptr<TensorOp> ResizeOperation::Build() {
// If size is a single value, the smaller edge of the image will be
// resized to this value with the same image aspect ratio.
int32_t height = size_[0];
int32_t width = 0;
@ -1730,6 +1816,115 @@ std::shared_ptr<TensorOp> RgbaToRgbOperation::Build() {
return tensor_op;
}
// SoftDvppDecodeRandomCropResizeJpegOperation
SoftDvppDecodeRandomCropResizeJpegOperation::SoftDvppDecodeRandomCropResizeJpegOperation(std::vector<int32_t> size,
std::vector<float> scale,
std::vector<float> ratio,
int32_t max_attempts)
: size_(size), scale_(scale), ratio_(ratio), max_attempts_(max_attempts) {}
Status SoftDvppDecodeRandomCropResizeJpegOperation::ValidateParams() {
// size
if (size_.size() != 2 && size_.size() != 1) {
std::string err_msg = "SoftDvppDecodeRandomCropResizeJpeg: size must be a vector of one or two values, got: " +
std::to_string(size_.size());
MS_LOG(ERROR) << err_msg;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
if (size_[0] <= 0 || (size_.size() == 2 && size_[1] <= 0)) {
std::string err_msg = "SoftDvppDecodeRandomCropResizeJpeg: size must only contain positive integers.";
MS_LOG(ERROR) << "SoftDvppDecodeRandomCropResizeJpeg: size must only contain positive integers, got: " << size_;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
// scale
if (scale_.size() != 2) {
std::string err_msg =
"SoftDvppDecodeRandomCropResizeJpeg: scale must be a vector of two values, got: " + std::to_string(scale_.size());
MS_LOG(ERROR) << err_msg;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
if (scale_[0] < 0 || scale_[1] < 0) {
std::string err_msg = "SoftDvppDecodeRandomCropResizeJpeg: scale must be greater than or equal to 0.";
MS_LOG(ERROR) << "SoftDvppDecodeRandomCropResizeJpeg: scale must be greater than or equal to 0, got: " << scale_;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
if (scale_[1] < scale_[0]) {
std::string err_msg = "SoftDvppDecodeRandomCropResizeJpeg: scale must be in the format of (min, max).";
MS_LOG(ERROR) << "SoftDvppDecodeRandomCropResizeJpeg: scale must be in the format of (min, max), but got: "
<< scale_;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
// ratio
if (ratio_.size() != 2) {
std::string err_msg =
"SoftDvppDecodeRandomCropResizeJpeg: ratio must be a vector of two values, got: " + std::to_string(ratio_.size());
MS_LOG(ERROR) << err_msg;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
if (ratio_[0] < 0 || ratio_[1] < 0) {
std::string err_msg = "SoftDvppDecodeRandomCropResizeJpeg: ratio must be greater than or equal to 0.";
MS_LOG(ERROR) << "SoftDvppDecodeRandomCropResizeJpeg: ratio must be greater than or equal to 0, got: " << ratio_;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
if (ratio_[1] < ratio_[0]) {
std::string err_msg = "SoftDvppDecodeRandomCropResizeJpeg: ratio must be in the format of (min, max).";
MS_LOG(ERROR) << "SoftDvppDecodeRandomCropResizeJpeg: ratio must be in the format of (min, max), but got: "
<< ratio_;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
// max_attempts
if (max_attempts_ < 1) {
std::string err_msg = "SoftDvppDecodeRandomCropResizeJpeg: max_attempts must be greater than or equal to 1, got: " +
std::to_string(max_attempts_);
MS_LOG(ERROR) << err_msg;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
return Status::OK();
}
std::shared_ptr<TensorOp> SoftDvppDecodeRandomCropResizeJpegOperation::Build() {
int32_t height = size_[0];
int32_t width = size_[0];
// User specified the width value.
if (size_.size() == 2) {
width = size_[1];
}
auto tensor_op = std::make_shared<SoftDvppDecodeRandomCropResizeJpegOp>(height, width, scale_[0], scale_[1],
ratio_[0], ratio_[1], max_attempts_);
return tensor_op;
}
// SoftDvppDecodeResizeJpegOperation
SoftDvppDecodeResizeJpegOperation::SoftDvppDecodeResizeJpegOperation(std::vector<int32_t> size) : size_(size) {}
Status SoftDvppDecodeResizeJpegOperation::ValidateParams() {
// size
if (size_.empty() || size_.size() > 2) {
std::string err_msg =
"SoftDvppDecodeResizeJpeg: size must be a vector of one or two values, got: " + std::to_string(size_.size());
MS_LOG(ERROR) << err_msg;
RETURN_STATUS_SYNTAX_ERROR(err_msg);
}
RETURN_IF_NOT_OK(ValidateVectorPositive("SoftDvppDecodeResizeJpeg", size_));
return Status::OK();
}
std::shared_ptr<TensorOp> SoftDvppDecodeResizeJpegOperation::Build() {
// If size is a single value, the smaller edge of the image will be
// resized to this value with the same image aspect ratio.
int32_t height = size_[0];
int32_t width = 0;
// User specified the width value.
if (size_.size() == 2) {
width = size_[1];
}
std::shared_ptr<SoftDvppDecodeResizeJpegOp> tensor_op = std::make_shared<SoftDvppDecodeResizeJpegOp>(height, width);
return tensor_op;
}
// SwapRedBlueOperation.
SwapRedBlueOperation::SwapRedBlueOperation() {}

151
mindspore/ccsrc/minddata/dataset/include/vision.h
View File

@ -17,7 +17,9 @@
#ifndef MINDSPORE_CCSRC_MINDDATA_DATASET_INCLUDE_VISION_H_
#define MINDSPORE_CCSRC_MINDDATA_DATASET_INCLUDE_VISION_H_
#include <map>
#include <memory>
#include <utility>
#include <vector>
#include "minddata/dataset/core/constants.h"
#include "minddata/dataset/include/transforms.h"
@ -42,6 +44,7 @@ class CutOutOperation;
#endif
class DecodeOperation;
#ifndef ENABLE_ANDROID
class EqualizeOperation;
class HwcToChwOperation;
class InvertOperation;
class MixUpBatchOperation;
@ -58,8 +61,11 @@ class RandomCropWithBBoxOperation;
class RandomHorizontalFlipOperation;
class RandomHorizontalFlipWithBBoxOperation;
class RandomPosterizeOperation;
class RandomResizeOperation;
class RandomResizeWithBBoxOperation;
class RandomResizedCropOperation;
class RandomRotationOperation;
class RandomSelectSubpolicyOperation;
class RandomSharpnessOperation;
class RandomSolarizeOperation;
class RandomVerticalFlipOperation;
@ -71,6 +77,8 @@ class ResizeOperation;
class ResizeWithBBoxOperation;
class RgbaToBgrOperation;
class RgbaToRgbOperation;
class SoftDvppDecodeRandomCropResizeJpegOperation;
class SoftDvppDecodeResizeJpegOperation;
class SwapRedBlueOperation;
class UniformAugOperation;
@ -129,6 +137,12 @@ std::shared_ptr<CutOutOperation> CutOut(int32_t length, int32_t num_patches = 1)
std::shared_ptr<DecodeOperation> Decode(bool rgb = true);
#ifndef ENABLE_ANDROID
/// \brief Function to create a Equalize TensorOperation.
/// \notes Apply histogram equalization on input image.
/// \return Shared pointer to the current TensorOperation.
std::shared_ptr<EqualizeOperation> Equalize();
/// \brief Function to create a HwcToChw TensorOperation.
/// \notes Transpose the input image; shape (H, W, C) to shape (C, H, W).
/// \return Shared pointer to the current TensorOperation.
@ -296,6 +310,21 @@ std::shared_ptr<RandomHorizontalFlipWithBBoxOperation> RandomHorizontalFlipWithB
/// \return Shared pointer to the current TensorOperation.
std::shared_ptr<RandomPosterizeOperation> RandomPosterize(const std::vector<uint8_t> &bit_range = {4, 8});
/// \brief Function to create a RandomResize TensorOperation.
/// \notes Resize the input image using a randomly selected interpolation mode.
/// \param[in] size A vector representing the output size of the resized image.
/// If size is a single value, the smaller edge of the image will be resized to this value with
// the same image aspect ratio. If size has 2 values, it should be (height, width).
std::shared_ptr<RandomResizeOperation> RandomResize(std::vector<int32_t> size);
/// \brief Function to create a RandomResizeWithBBox TensorOperation.
/// \notes Resize the input image using a randomly selected interpolation mode and adjust
/// bounding boxes accordingly.
/// \param[in] size A vector representing the output size of the resized image.
/// If size is a single value, the smaller edge of the image will be resized to this value with
// the same image aspect ratio. If size has 2 values, it should be (height, width).
std::shared_ptr<RandomResizeWithBBoxOperation> RandomResizeWithBBox(std::vector<int32_t> size);
/// \brief Function to create a RandomResizedCrop TensorOperation.
/// \notes Crop the input image to a random size and aspect ratio.
/// \param[in] size A vector representing the output size of the cropped image.
@ -325,6 +354,15 @@ std::shared_ptr<RandomRotationOperation> RandomRotation(
std::vector<float> degrees, InterpolationMode resample = InterpolationMode::kNearestNeighbour, bool expand = false,
std::vector<float> center = {-1, -1}, std::vector<uint8_t> fill_value = {0, 0, 0});
/// \brief Function to create a RandomSelectSubpolicy TensorOperation.
/// \notes Choose a random sub-policy from a list to be applied on the input image. A sub-policy is a list of tuples
/// (op, prob), where op is a TensorOp operation and prob is the probability that this op will be applied. Once
/// a sub-policy is selected, each op within the subpolicy with be applied in sequence according to its probability.
/// \param[in] policy Vector of sub-policies to choose from.
/// \return Shared pointer to the current TensorOperation.
std::shared_ptr<RandomSelectSubpolicyOperation> RandomSelectSubpolicy(
std::vector<std::vector<std::pair<std::shared_ptr<TensorOperation>, double>>> policy);
/// \brief Function to create a RandomSharpness TensorOperation.
/// \notes Tensor operation to perform random sharpness.
/// \param[in] degrees A float vector of size 2, representing the starting and ending degree to uniformly
@ -390,6 +428,35 @@ std::shared_ptr<RgbaToBgrOperation> RGBA2BGR();
/// \return Shared pointer to the current TensorOperation.
std::shared_ptr<RgbaToRgbOperation> RGBA2RGB();
/// \brief Function to create a SoftDvppDecodeRandomCropResizeJpeg TensorOperation.
/// \notes Tensor operation to decode, random crop and resize JPEG image using the simulation algorithm of
/// Ascend series chip DVPP module. The usage scenario is consistent with SoftDvppDecodeResizeJpeg.
/// The input image size should be in range [32*32, 8192*8192].
/// The zoom-out and zoom-in multiples of the image length and width should in the range [1/32, 16].
/// Only images with an even resolution can be output. The output of odd resolution is not supported.
/// \param[in] size A vector representing the output size of the resized image.
/// If size is a single value, smaller edge of the image will be resized to this value with
/// the same image aspect ratio. If size has 2 values, it should be (height, width).
/// \return Shared pointer to the current TensorOperation.
std::shared_ptr<SoftDvppDecodeRandomCropResizeJpegOperation> SoftDvppDecodeRandomCropResizeJpeg(
std::vector<int32_t> size, std::vector<float> scale = {0.08, 1.0}, std::vector<float> ratio = {3. / 4., 4. / 3.},
int32_t max_attempts = 10);
/// \brief Function to create a SoftDvppDecodeResizeJpeg TensorOperation.
/// \notes Tensor operation to decode and resize JPEG image using the simulation algorithm of Ascend series
/// chip DVPP module. It is recommended to use this algorithm in the following scenarios:
/// When training, the DVPP of the Ascend chip is not used,
/// and the DVPP of the Ascend chip is used during inference,
/// and the accuracy of inference is lower than the accuracy of training;
/// and the input image size should be in range [32*32, 8192*8192].
/// The zoom-out and zoom-in multiples of the image length and width should in the range [1/32, 16].
/// Only images with an even resolution can be output. The output of odd resolution is not supported.
/// \param[in] size A vector representing the output size of the resized image.
/// If size is a single value, smaller edge of the image will be resized to this value with
/// the same image aspect ratio. If size has 2 values, it should be (height, width).
/// \return Shared pointer to the current TensorOperation.
std::shared_ptr<SoftDvppDecodeResizeJpegOperation> SoftDvppDecodeResizeJpeg(std::vector<int32_t> size);
/// \brief Function to create a SwapRedBlue TensorOp
/// \notes Swaps the red and blue channels in image
/// \return Shared pointer to the current TensorOp
@ -512,6 +579,15 @@ class DecodeOperation : public TensorOperation {
};
#ifndef ENABLE_ANDROID
class EqualizeOperation : public TensorOperation {
public:
~EqualizeOperation() = default;
std::shared_ptr<TensorOp> Build() override;
Status ValidateParams() override;
};
class HwcToChwOperation : public TensorOperation {
public:
~HwcToChwOperation() = default;
@ -735,6 +811,34 @@ class RandomPosterizeOperation : public TensorOperation {
std::vector<uint8_t> bit_range_;
};
class RandomResizeOperation : public TensorOperation {
public:
explicit RandomResizeOperation(std::vector<int32_t> size);
~RandomResizeOperation() = default;
std::shared_ptr<TensorOp> Build() override;
Status ValidateParams() override;
private:
std::vector<int32_t> size_;
};
class RandomResizeWithBBoxOperation : public TensorOperation {
public:
explicit RandomResizeWithBBoxOperation(std::vector<int32_t> size);
~RandomResizeWithBBoxOperation() = default;
std::shared_ptr<TensorOp> Build() override;
Status ValidateParams() override;
private:
std::vector<int32_t> size_;
};
class RandomResizedCropOperation : public TensorOperation {
public:
explicit RandomResizedCropOperation(std::vector<int32_t> size, std::vector<float> scale = {0.08, 1.0},
@ -775,6 +879,21 @@ class RandomRotationOperation : public TensorOperation {
std::vector<uint8_t> fill_value_;
};
class RandomSelectSubpolicyOperation : public TensorOperation {
public:
explicit RandomSelectSubpolicyOperation(
std::vector<std::vector<std::pair<std::shared_ptr<TensorOperation>, double>>> policy);
~RandomSelectSubpolicyOperation() = default;
std::shared_ptr<TensorOp> Build() override;
Status ValidateParams() override;
private:
std::vector<std::vector<std::pair<std::shared_ptr<TensorOperation>, double>>> policy_;
};
class RandomSharpnessOperation : public TensorOperation {
public:
explicit RandomSharpnessOperation(std::vector<float> degrees = {0.1, 1.9});
@ -902,6 +1021,38 @@ class RgbaToRgbOperation : public TensorOperation {
Status ValidateParams() override;
};
class SoftDvppDecodeRandomCropResizeJpegOperation : public TensorOperation {
public:
explicit SoftDvppDecodeRandomCropResizeJpegOperation(std::vector<int32_t> size, std::vector<float> scale,
std::vector<float> ratio, int32_t max_attempts);
~SoftDvppDecodeRandomCropResizeJpegOperation() = default;
std::shared_ptr<TensorOp> Build() override;
Status ValidateParams() override;
private:
std::vector<int32_t> size_;
std::vector<float> scale_;
std::vector<float> ratio_;
int32_t max_attempts_;
};
class SoftDvppDecodeResizeJpegOperation : public TensorOperation {
public:
explicit SoftDvppDecodeResizeJpegOperation(std::vector<int32_t> size);
~SoftDvppDecodeResizeJpegOperation() = default;
std::shared_ptr<TensorOp> Build() override;
Status ValidateParams() override;
private:
std::vector<int32_t> size_;
};
class SwapRedBlueOperation : public TensorOperation {
public:
SwapRedBlueOperation();

2
mindspore/dataset/vision/c_transforms.py
View File

@ -1260,7 +1260,7 @@ class SoftDvppDecodeRandomCropResizeJpeg(cde.SoftDvppDecodeRandomCropResizeJpegO
Tensor operation to decode, random crop and resize JPEG image using the simulation algorithm of
Ascend series chip DVPP module.
The usage scenario is consistent with SoftDvppDecodeReiszeJpeg.
The usage scenario is consistent with SoftDvppDecodeResizeJpeg.
The input image size should be in range [32*32, 8192*8192].
The zoom-out and zoom-in multiples of the image length and width should in the range [1/32, 16].
Only images with an even resolution can be output. The output of odd resolution is not supported.

502
tests/ut/cpp/dataset/c_api_vision_test.cc
View File

@ -1640,6 +1640,208 @@ TEST_F(MindDataTestPipeline, TestRandomPosterizeSuccess2) {
iter->Stop();
}
TEST_F(MindDataTestPipeline, TestRandomResizeSuccess1) {
MS_LOG(INFO) << "Doing MindDataTestPipeline-TestRandomResizeSuccess1 with single integer input.";
// Create an ImageFolder Dataset
std::string folder_path = datasets_root_path_ + "/testPK/data/";
std::shared_ptr<Dataset> ds = ImageFolder(folder_path, true, RandomSampler(false, 5));
EXPECT_NE(ds, nullptr);
// Create objects for the tensor ops
std::shared_ptr<TensorOperation> random_resize = vision::RandomResize({66});
EXPECT_NE(random_resize, nullptr);
// Create a Map operation on ds
ds = ds->Map({random_resize}, {"image"});
EXPECT_NE(ds, nullptr);
// Create an iterator over the result of the above dataset
// This will trigger the creation of the Execution Tree and launch it.
std::shared_ptr<Iterator> iter = ds->CreateIterator();
EXPECT_NE(iter, nullptr);
// Iterate the dataset and get each row
std::unordered_map<std::string, std::shared_ptr<Tensor>> row;
iter->GetNextRow(&row);
uint64_t i = 0;
while (row.size() != 0) {
i++;
auto image = row["image"];
MS_LOG(INFO) << "Tensor image shape: " << image->shape();
EXPECT_EQ(image->shape()[0] == 66, true);
iter->GetNextRow(&row);
}
EXPECT_EQ(i, 5);
// Manually terminate the pipeline
iter->Stop();
}
TEST_F(MindDataTestPipeline, TestRandomResizeSuccess2) {
MS_LOG(INFO) << "Doing MindDataTestPipeline-TestRandomResizeSuccess2 with (height, width) input.";
// Create an ImageFolder Dataset
std::string folder_path = datasets_root_path_ + "/testPK/data/";
std::shared_ptr<Dataset> ds = ImageFolder(folder_path, true, RandomSampler(false, 3));
EXPECT_NE(ds, nullptr);
// Create a Repeat operation on ds
int32_t repeat_num = 2;
ds = ds->Repeat(repeat_num);
EXPECT_NE(ds, nullptr);
// Create objects for the tensor ops
std::shared_ptr<TensorOperation> random_resize = vision::RandomResize({66, 77});
EXPECT_NE(random_resize, nullptr);
// Create a Map operation on ds
ds = ds->Map({random_resize}, {"image"});
EXPECT_NE(ds, nullptr);
// Create an iterator over the result of the above dataset
// This will trigger the creation of the Execution Tree and launch it.
std::shared_ptr<Iterator> iter = ds->CreateIterator();
EXPECT_NE(iter, nullptr);
// Iterate the dataset and get each row
std::unordered_map<std::string, std::shared_ptr<Tensor>> row;
iter->GetNextRow(&row);
uint64_t i = 0;
while (row.size() != 0) {
i++;
auto image = row["image"];
MS_LOG(INFO) << "Tensor image shape: " << image->shape();
EXPECT_EQ(image->shape()[0] == 66 && image->shape()[1] == 77, true);
iter->GetNextRow(&row);
}
EXPECT_EQ(i, 6);
// Manually terminate the pipeline
iter->Stop();
}
TEST_F(MindDataTestPipeline, TestRandomResizeFail) {
MS_LOG(INFO) << "Doing MindDataTestPipeline-TestRandomResizeFail incorrect size.";
// RandomResize : size must only contain positive integers
std::shared_ptr<TensorOperation> random_resize1 = vision::RandomResize({-66, 77});
EXPECT_EQ(random_resize1, nullptr);
// RandomResize : size must be a vector of one or two values
std::shared_ptr<TensorOperation> random_resize2 = vision::RandomResize({1, 2, 3});
EXPECT_EQ(random_resize2, nullptr);
// RandomResize : size must be a vector of one or two values
std::shared_ptr<TensorOperation> random_resize3 = vision::RandomResize({});
EXPECT_EQ(random_resize3, nullptr);
}
TEST_F(MindDataTestPipeline, TestRandomResizeWithBBoxSuccess1) {
MS_LOG(INFO) << "Doing MindDataTestPipeline-TestRandomResizeWithBBoxSuccess1 with single integer input.";
// Create an VOC Dataset
std::string folder_path = datasets_root_path_ + "/testVOC2012_2";
std::shared_ptr<Dataset> ds = VOC(folder_path, "Detection", "train", {}, true, SequentialSampler(0, 3));
EXPECT_NE(ds, nullptr);
// Create objects for the tensor ops
std::shared_ptr<TensorOperation> random_resize = vision::RandomResizeWithBBox({88});
EXPECT_NE(random_resize, nullptr);
// Create a Map operation on ds
ds = ds->Map({random_resize}, {"image", "bbox"});
EXPECT_NE(ds, nullptr);
// Create an iterator over the result of the above dataset
// This will trigger the creation of the Execution Tree and launch it.
std::shared_ptr<Iterator> iter = ds->CreateIterator();
EXPECT_NE(iter, nullptr);
// Iterate the dataset and get each row
std::unordered_map<std::string, std::shared_ptr<Tensor>> row;
iter->GetNextRow(&row);
uint64_t i = 0;
while (row.size() != 0) {
i++;
auto image = row["image"];
MS_LOG(INFO) << "Tensor image shape: " << image->shape();
EXPECT_EQ(image->shape()[0] == 88, true);
iter->GetNextRow(&row);
}
EXPECT_EQ(i, 3);
// Manually terminate the pipeline
iter->Stop();
}
TEST_F(MindDataTestPipeline, TestRandomResizeWithBBoxSuccess2) {
MS_LOG(INFO) << "Doing MindDataTestPipeline-TestRandomResizeWithBBoxSuccess2 with (height, width) input.";
// Create an VOC Dataset
std::string folder_path = datasets_root_path_ + "/testVOC2012_2";
std::shared_ptr<Dataset> ds = VOC(folder_path, "Detection", "train", {}, true, SequentialSampler(0, 4));
EXPECT_NE(ds, nullptr);
// Create a Repeat operation on ds
int32_t repeat_num = 2;
ds = ds->Repeat(repeat_num);
EXPECT_NE(ds, nullptr);
// Create objects for the tensor ops
std::shared_ptr<TensorOperation> random_resize = vision::RandomResizeWithBBox({88, 99});
EXPECT_NE(random_resize, nullptr);
// Create a Map operation on ds
ds = ds->Map({random_resize}, {"image", "bbox"});
EXPECT_NE(ds, nullptr);
// Create an iterator over the result of the above dataset
// This will trigger the creation of the Execution Tree and launch it.
std::shared_ptr<Iterator> iter = ds->CreateIterator();
EXPECT_NE(iter, nullptr);
// Iterate the dataset and get each row
std::unordered_map<std::string, std::shared_ptr<Tensor>> row;
iter->GetNextRow(&row);
uint64_t i = 0;
while (row.size() != 0) {
i++;
auto image = row["image"];
MS_LOG(INFO) << "Tensor image shape: " << image->shape();
EXPECT_EQ(image->shape()[0] == 88 && image->shape()[1] == 99, true);
iter->GetNextRow(&row);
}
EXPECT_EQ(i, 8);
// Manually terminate the pipeline
iter->Stop();
}
TEST_F(MindDataTestPipeline, TestRandomResizeWithBBoxFail) {
MS_LOG(INFO) << "Doing MindDataTestPipeline-TestRandomResizeWithBBoxFail incorrect size.";
// RandomResizeWithBBox : size must only contain positive integers
std::shared_ptr<TensorOperation> random_resize_with_bbox1 = vision::RandomResizeWithBBox({-66, 77});
EXPECT_EQ(random_resize_with_bbox1, nullptr);
// RandomResizeWithBBox : size must be a vector of one or two values
std::shared_ptr<TensorOperation> random_resize_with_bbox2 = vision::RandomResizeWithBBox({1, 2, 3});
EXPECT_EQ(random_resize_with_bbox2, nullptr);
// RandomResizeWithBBox : size must be a vector of one or two values
std::shared_ptr<TensorOperation> random_resize_with_bbox3 = vision::RandomResizeWithBBox({});
EXPECT_EQ(random_resize_with_bbox3, nullptr);
}
TEST_F(MindDataTestPipeline, TestRandomResizedCropSuccess1) {
// Testing RandomResizedCrop with default values
// Create a Cifar10 Dataset
@ -1864,6 +2066,75 @@ TEST_F(MindDataTestPipeline, TestRandomRotationFail) {
EXPECT_EQ(random_rotation_op6, nullptr);
}
TEST_F(MindDataTestPipeline, TestRandomSelectSubpolicySuccess) {
MS_LOG(INFO) << "Doing MindDataTestPipeline-TestRandomSelectSubpolicySuccess.";
// Create an ImageFolder Dataset
std::string folder_path = datasets_root_path_ + "/testPK/data/";
std::shared_ptr<Dataset> ds = ImageFolder(folder_path, true, RandomSampler(false, 7));
EXPECT_NE(ds, nullptr);
// Create objects for the tensor ops
// Valid case: TensorOperation is not null and probability is between (0,1)
std::shared_ptr<TensorOperation> random_select_subpolicy = vision::RandomSelectSubpolicy(
{{{vision::Invert(), 0.5}, {vision::Equalize(), 0.5}}, {{vision::Resize({15, 15}), 1}}});
EXPECT_NE(random_select_subpolicy, nullptr);
// Create a Map operation on ds
ds = ds->Map({random_select_subpolicy});
EXPECT_NE(ds, nullptr);
// Create an iterator over the result of the above dataset
// This will trigger the creation of the Execution Tree and launch it.
std::shared_ptr<Iterator> iter = ds->CreateIterator();
EXPECT_NE(iter, nullptr);
// Iterate the dataset and get each row
std::unordered_map<std::string, std::shared_ptr<Tensor>> row;
iter->GetNextRow(&row);
uint64_t i = 0;
while (row.size() != 0) {
i++;
auto image = row["image"];
MS_LOG(INFO) << "Tensor image shape: " << image->shape();
iter->GetNextRow(&row);
}
EXPECT_EQ(i, 7);
// Manually terminate the pipeline
iter->Stop();
}
TEST_F(MindDataTestPipeline, TestRandomSelectSubpolicyFail) {
MS_LOG(INFO) << "Doing MindDataTestPipeline-TestRandomSelectSubpolicyFail.";
// RandomSelectSubpolicy : probability of transform must be between 0.0 and 1.0
std::shared_ptr<TensorOperation> random_select_subpolicy1 = vision::RandomSelectSubpolicy(
{{{vision::Invert(), 1.5}, {vision::Equalize(), 0.5}}, {{vision::Resize({15, 15}), 1}}});
EXPECT_EQ(random_select_subpolicy1, nullptr);
// RandomSelectSubpolicy: policy must not be empty
std::shared_ptr<TensorOperation> random_select_subpolicy2 = vision::RandomSelectSubpolicy(
{{{vision::Invert(), 0.5}, {vision::Equalize(), 0.5}}, {{nullptr, 1}}});
EXPECT_EQ(random_select_subpolicy2, nullptr);
// RandomSelectSubpolicy: policy must not be empty
std::shared_ptr<TensorOperation> random_select_subpolicy3 = vision::RandomSelectSubpolicy({});
EXPECT_EQ(random_select_subpolicy3, nullptr);
// RandomSelectSubpolicy: policy must not be empty
std::shared_ptr<TensorOperation> random_select_subpolicy4 = vision::RandomSelectSubpolicy(
{{{vision::Invert(), 0.5}, {vision::Equalize(), 0.5}}, {}});
EXPECT_EQ(random_select_subpolicy4, nullptr);
// RandomSelectSubpolicy: policy must not be empty
std::shared_ptr<TensorOperation> random_select_subpolicy5 = vision::RandomSelectSubpolicy(
{{{}, {vision::Equalize(), 0.5}}, {{vision::Resize({15, 15}), 1}}});
EXPECT_EQ(random_select_subpolicy5, nullptr);
}
TEST_F(MindDataTestPipeline, TestRandomSharpness) {
MS_LOG(INFO) << "Doing MindDataTestPipeline-TestRandomSharpness.";
@ -2297,6 +2568,237 @@ TEST_F(MindDataTestPipeline, TestRescaleFail) {
EXPECT_EQ(rescale, nullptr);
}
TEST_F(MindDataTestPipeline, TestSoftDvppDecodeRandomCropResizeJpegSuccess1) {
MS_LOG(INFO) << "Doing MindDataTestPipeline-TestSoftDvppDecodeRandomCropResizeJpegSuccess1 with single integer input.";
// Create an ImageFolder Dataset
std::string folder_path = datasets_root_path_ + "/testPK/data/";
std::shared_ptr<Dataset> ds = ImageFolder(folder_path, false, RandomSampler(false, 4));
EXPECT_NE(ds, nullptr);
// Create objects for the tensor ops
std::shared_ptr<TensorOperation> soft_dvpp_decode_random_crop_resize_jpeg =
vision::SoftDvppDecodeRandomCropResizeJpeg({500});
EXPECT_NE(soft_dvpp_decode_random_crop_resize_jpeg, nullptr);
// Create a Map operation on ds
ds = ds->Map({soft_dvpp_decode_random_crop_resize_jpeg}, {"image"});
EXPECT_NE(ds, nullptr);
// Create an iterator over the result of the above dataset
// This will trigger the creation of the Execution Tree and launch it.
std::shared_ptr<Iterator> iter = ds->CreateIterator();
EXPECT_NE(iter, nullptr);
// Iterate the dataset and get each row
std::unordered_map<std::string, std::shared_ptr<Tensor>> row;
iter->GetNextRow(&row);
uint64_t i = 0;
while (row.size() != 0) {
i++;
auto image = row["image"];
MS_LOG(INFO) << "Tensor image shape: " << image->shape();
EXPECT_EQ(image->shape()[0] == 500 && image->shape()[1] == 500, true);
iter->GetNextRow(&row);
}
EXPECT_EQ(i, 4);
// Manually terminate the pipeline
iter->Stop();
}
TEST_F(MindDataTestPipeline, TestSoftDvppDecodeRandomCropResizeJpegSuccess2) {
MS_LOG(INFO) << "Doing MindDataTestPipeline-TestSoftDvppDecodeRandomCropResizeJpegSuccess2 with (height, width) input.";
// Create an ImageFolder Dataset
std::string folder_path = datasets_root_path_ + "/testPK/data/";
std::shared_ptr<Dataset> ds = ImageFolder(folder_path, false, RandomSampler(false, 6));
EXPECT_NE(ds, nullptr);
// Create objects for the tensor ops
std::shared_ptr<TensorOperation> soft_dvpp_decode_random_crop_resize_jpeg =
vision::SoftDvppDecodeRandomCropResizeJpeg({500, 600}, {0.25, 0.75}, {0.5, 1.25}, 20);
EXPECT_NE(soft_dvpp_decode_random_crop_resize_jpeg, nullptr);
// Create a Map operation on ds
ds = ds->Map({soft_dvpp_decode_random_crop_resize_jpeg}, {"image"});
EXPECT_NE(ds, nullptr);
// Create an iterator over the result of the above dataset
// This will trigger the creation of the Execution Tree and launch it.
std::shared_ptr<Iterator> iter = ds->CreateIterator();
EXPECT_NE(iter, nullptr);
// Iterate the dataset and get each row
std::unordered_map<std::string, std::shared_ptr<Tensor>> row;
iter->GetNextRow(&row);
uint64_t i = 0;
while (row.size() != 0) {
i++;
auto image = row["image"];
MS_LOG(INFO) << "Tensor image shape: " << image->shape();
EXPECT_EQ(image->shape()[0] == 500 && image->shape()[1] == 600, true);
iter->GetNextRow(&row);
}
EXPECT_EQ(i, 6);
// Manually terminate the pipeline
iter->Stop();
}
TEST_F(MindDataTestPipeline, TestSoftDvppDecodeRandomCropResizeJpegFail) {
MS_LOG(INFO) << "Doing MindDataTestPipeline-TestSoftDvppDecodeRandomCropResizeJpegFail with incorrect parameters.";
// SoftDvppDecodeRandomCropResizeJpeg: size must only contain positive integers
auto soft_dvpp_decode_random_crop_resize_jpeg1 = vision::SoftDvppDecodeRandomCropResizeJpeg({-500, 600});
EXPECT_EQ(soft_dvpp_decode_random_crop_resize_jpeg1, nullptr);
// SoftDvppDecodeRandomCropResizeJpeg: size must only contain positive integers
auto soft_dvpp_decode_random_crop_resize_jpeg2 = vision::SoftDvppDecodeRandomCropResizeJpeg({-500});
EXPECT_EQ(soft_dvpp_decode_random_crop_resize_jpeg2, nullptr);
// SoftDvppDecodeRandomCropResizeJpeg: size must be a vector of one or two values
auto soft_dvpp_decode_random_crop_resize_jpeg3 = vision::SoftDvppDecodeRandomCropResizeJpeg({500, 600, 700});
EXPECT_EQ(soft_dvpp_decode_random_crop_resize_jpeg3, nullptr);
// SoftDvppDecodeRandomCropResizeJpeg: scale must be greater than or equal to 0
auto soft_dvpp_decode_random_crop_resize_jpeg4 = vision::SoftDvppDecodeRandomCropResizeJpeg({500}, {-0.1, 0.9});
EXPECT_EQ(soft_dvpp_decode_random_crop_resize_jpeg4, nullptr);
// SoftDvppDecodeRandomCropResizeJpeg: scale must be in the format of (min, max)
auto soft_dvpp_decode_random_crop_resize_jpeg5 = vision::SoftDvppDecodeRandomCropResizeJpeg({500}, {0.6, 0.2});
EXPECT_EQ(soft_dvpp_decode_random_crop_resize_jpeg5, nullptr);
// SoftDvppDecodeRandomCropResizeJpeg: scale must be a vector of two values
auto soft_dvpp_decode_random_crop_resize_jpeg6 = vision::SoftDvppDecodeRandomCropResizeJpeg({500}, {0.5, 0.6, 0.7});
EXPECT_EQ(soft_dvpp_decode_random_crop_resize_jpeg6, nullptr);
// SoftDvppDecodeRandomCropResizeJpeg: ratio must be greater than or equal to 0
auto soft_dvpp_decode_random_crop_resize_jpeg7 =
vision::SoftDvppDecodeRandomCropResizeJpeg({500}, {0.5, 0.9}, {-0.2, 0.4});
EXPECT_EQ(soft_dvpp_decode_random_crop_resize_jpeg7, nullptr);
// SoftDvppDecodeRandomCropResizeJpeg: ratio must be in the format of (min, max)
auto soft_dvpp_decode_random_crop_resize_jpeg8 =
vision::SoftDvppDecodeRandomCropResizeJpeg({500}, {0.5, 0.9}, {0.4, 0.2});
EXPECT_EQ(soft_dvpp_decode_random_crop_resize_jpeg8, nullptr);
// SoftDvppDecodeRandomCropResizeJpeg: ratio must be a vector of two values
auto soft_dvpp_decode_random_crop_resize_jpeg9 =
vision::SoftDvppDecodeRandomCropResizeJpeg({500}, {0.5, 0.9}, {0.1, 0.2, 0.3});
EXPECT_EQ(soft_dvpp_decode_random_crop_resize_jpeg9, nullptr);
// SoftDvppDecodeRandomCropResizeJpeg: max_attempts must be greater than or equal to 1
auto soft_dvpp_decode_random_crop_resize_jpeg10 =
vision::SoftDvppDecodeRandomCropResizeJpeg({500}, {0.5, 0.9}, {0.1, 0.2}, 0);
EXPECT_EQ(soft_dvpp_decode_random_crop_resize_jpeg10, nullptr);
}
TEST_F(MindDataTestPipeline, TestSoftDvppDecodeResizeJpegSuccess1) {
MS_LOG(INFO) << "Doing MindDataTestPipeline-TestSoftDvppDecodeResizeJpegSuccess1 with single integer input.";
// Create an ImageFolder Dataset
std::string folder_path = datasets_root_path_ + "/testPK/data/";
std::shared_ptr<Dataset> ds = ImageFolder(folder_path, false, RandomSampler(false, 4));
EXPECT_NE(ds, nullptr);
// Create a Repeat operation on ds
int32_t repeat_num = 3;
ds = ds->Repeat(repeat_num);
EXPECT_NE(ds, nullptr);
// Create SoftDvppDecodeResizeJpeg object with single integer input
std::shared_ptr<TensorOperation> soft_dvpp_decode_resize_jpeg_op = vision::SoftDvppDecodeResizeJpeg({1134});
EXPECT_NE(soft_dvpp_decode_resize_jpeg_op, nullptr);
// Create a Map operation on ds
ds = ds->Map({soft_dvpp_decode_resize_jpeg_op});
EXPECT_NE(ds, nullptr);
// Create an iterator over the result of the above dataset
// This will trigger the creation of the Execution Tree and launch it.
std::shared_ptr<Iterator> iter = ds->CreateIterator();
EXPECT_NE(iter, nullptr);
// Iterate the dataset and get each row
std::unordered_map<std::string, std::shared_ptr<Tensor>> row;
iter->GetNextRow(&row);
uint64_t i = 0;
while (row.size() != 0) {
i++;
auto image = row["image"];
MS_LOG(INFO) << "Tensor image shape: " << image->shape();
iter->GetNextRow(&row);
}
EXPECT_EQ(i, 12);
// Manually terminate the pipeline
iter->Stop();
}
TEST_F(MindDataTestPipeline, TestSoftDvppDecodeResizeJpegSuccess2) {
MS_LOG(INFO) << "Doing MindDataTestPipeline-TestSoftDvppDecodeResizeJpegSuccess2 with (height, width) input.";
// Create an ImageFolder Dataset
std::string folder_path = datasets_root_path_ + "/testPK/data/";
std::shared_ptr<Dataset> ds = ImageFolder(folder_path, false, RandomSampler(false, 2));
EXPECT_NE(ds, nullptr);
// Create SoftDvppDecodeResizeJpeg object with single integer input
std::shared_ptr<TensorOperation> soft_dvpp_decode_resize_jpeg_op = vision::SoftDvppDecodeResizeJpeg({100, 200});
EXPECT_NE(soft_dvpp_decode_resize_jpeg_op, nullptr);
// Create a Map operation on ds
ds = ds->Map({soft_dvpp_decode_resize_jpeg_op});
EXPECT_NE(ds, nullptr);
// Create an iterator over the result of the above dataset
// This will trigger the creation of the Execution Tree and launch it.
std::shared_ptr<Iterator> iter = ds->CreateIterator();
EXPECT_NE(iter, nullptr);
// Iterate the dataset and get each row
std::unordered_map<std::string, std::shared_ptr<Tensor>> row;
iter->GetNextRow(&row);
uint64_t i = 0;
while (row.size() != 0) {
i++;
auto image = row["image"];
MS_LOG(INFO) << "Tensor image shape: " << image->shape();
iter->GetNextRow(&row);
}
EXPECT_EQ(i, 2);
// Manually terminate the pipeline
iter->Stop();
}
TEST_F(MindDataTestPipeline, TestSoftDvppDecodeResizeJpegFail) {
MS_LOG(INFO) << "Doing MindDataTestPipeline-TestSoftDvppDecodeResizeJpegFail with incorrect size.";
// CSoftDvppDecodeResizeJpeg: size must be a vector of one or two values
std::shared_ptr<TensorOperation> soft_dvpp_decode_resize_jpeg_op1 = vision::SoftDvppDecodeResizeJpeg({});
EXPECT_EQ(soft_dvpp_decode_resize_jpeg_op1, nullptr);
// SoftDvppDecodeResizeJpeg: size must be a vector of one or two values
std::shared_ptr<TensorOperation> soft_dvpp_decode_resize_jpeg_op2 = vision::SoftDvppDecodeResizeJpeg({1, 2, 3});
EXPECT_EQ(soft_dvpp_decode_resize_jpeg_op2, nullptr);
// SoftDvppDecodeResizeJpeg: size must only contain positive integers
std::shared_ptr<TensorOperation> soft_dvpp_decode_resize_jpeg_op3 = vision::SoftDvppDecodeResizeJpeg({20, -20});
EXPECT_EQ(soft_dvpp_decode_resize_jpeg_op3, nullptr);
// SoftDvppDecodeResizeJpeg: size must only contain positive integers
std::shared_ptr<TensorOperation> soft_dvpp_decode_resize_jpeg_op4 = vision::SoftDvppDecodeResizeJpeg({0});
EXPECT_EQ(soft_dvpp_decode_resize_jpeg_op4, nullptr);
}
TEST_F(MindDataTestPipeline, DISABLED_TestUniformAugmentFail1) {
MS_LOG(INFO) << "Doing MindDataTestPipeline-TestUniformAugmentFail1 with invalid zero num_ops parameter.";