Skip to content

[Part 2] Image Transformer #721

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 8 commits into
base: master
Choose a base branch
from
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
28 changes: 28 additions & 0 deletions tensorflow_lite_support/cc/task/vision/BUILD
Original file line number Diff line number Diff line change
Expand Up @@ -173,3 +173,31 @@ cc_library_with_tflite(
"@org_tensorflow//tensorflow/lite/core/api:op_resolver",
],
)

# IMPORTANT: in order to use hardware acceleration delegates, configurable through the
# `compute_settings` field of the ImageClassifierOptions, you must additionally link to
# the appropriate delegate plugin target (e.g. `gpu_plugin` for GPU) from:
# https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/experimental/acceleration/configuration/BUILD
# To use EDGETPU_CORAL, link to `edgetpu_coral_plugin` from:
# https://github.com/tensorflow/tflite-support/blob/a58a4f9225c411fa9ba29f821523e6e283988d23/tensorflow_lite_support/acceleration/configuration/BUILD#L11
cc_library_with_tflite(
name = "image_transformer",
srcs = ["image_transformer.cc"],
hdrs = ["image_transformer.h"],
tflite_deps = [
"@org_tensorflow//tensorflow/lite/core/shims:builtin_ops",
"//tensorflow_lite_support/cc/task/core:task_api_factory",
"//tensorflow_lite_support/cc/task/vision/core:base_vision_task_api",
"//tensorflow_lite_support/cc/task/processor:image_postprocessor",
],
deps = [
"//tensorflow_lite_support/cc/port:integral_types",
"//tensorflow_lite_support/cc/port:status_macros",
"//tensorflow_lite_support/cc/port:statusor",
"//tensorflow_lite_support/cc/task/core:external_file_handler",
"//tensorflow_lite_support/cc/task/vision/proto:image_transformer_options_proto_inc",
"@com_google_absl//absl/strings:str_format",
"@flatbuffers",
"@org_tensorflow//tensorflow/lite/core/api",
],
)
123 changes: 123 additions & 0 deletions tensorflow_lite_support/cc/task/vision/image_transformer.cc
Original file line number Diff line number Diff line change
@@ -0,0 +1,123 @@
/* Copyright 2021 The TensorFlow Authors. All Rights Reserved.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
==============================================================================*/

#include "tensorflow_lite_support/cc/task/vision/image_transformer.h"

#include "external/com_google_absl/absl/strings/str_format.h"
#include "external/com_google_absl/absl/strings/string_view.h"
#include "flatbuffers/flatbuffers.h" // from @flatbuffers
#include "tensorflow_lite_support/cc/port/status_macros.h"
#include "tensorflow_lite_support/cc/task/core/task_api_factory.h"

namespace tflite {
namespace task {
namespace vision {

namespace {

using ::absl::StatusCode;
using ::tflite::support::CreateStatusWithPayload;
using ::tflite::support::StatusOr;
using ::tflite::support::TfLiteSupportStatus;
using ::tflite::task::core::AssertAndReturnTypedTensor;
using ::tflite::task::core::TaskAPIFactory;
using ::tflite::task::core::TfLiteEngine;
using ::tflite::task::vision::FrameBuffer;
} // namespace

/* static */
StatusOr<std::unique_ptr<ImageTransformer>> ImageTransformer::CreateFromOptions(
const ImageTransformerOptions& options,
std::unique_ptr<tflite::OpResolver> resolver) {
RETURN_IF_ERROR(SanityCheckOptions(options));

// Copy options to ensure the ExternalFile outlives the constructed object.
auto options_copy = absl::make_unique<ImageTransformerOptions>(options);

std::unique_ptr<ImageTransformer> image_transformer;

ASSIGN_OR_RETURN(image_transformer,
TaskAPIFactory::CreateFromBaseOptions<ImageTransformer>(
&options_copy->base_options(), std::move(resolver)));

RETURN_IF_ERROR(image_transformer->Init(std::move(options_copy)));
return image_transformer;
}

/* static */
absl::Status ImageTransformer::SanityCheckOptions(
const ImageTransformerOptions& options) {
// Nothing to do.
return absl::OkStatus();
}

absl::Status ImageTransformer::Init(
std::unique_ptr<ImageTransformerOptions> options) {
// Set options.
options_ = std::move(options);

// Perform pre-initialization actions (by default, sets the process engine for
// image pre-processing to kLibyuv as a sane default).
RETURN_IF_ERROR(PreInit());

// Sanity check and set inputs and outputs.
RETURN_IF_ERROR(CheckAndSetInputs());
RETURN_IF_ERROR(CheckAndSetOutputs());

RETURN_IF_ERROR(PostInit());

ASSIGN_OR_RETURN(postprocessor_, processor::ImagePostprocessor::Create(
GetTfLiteEngine(), {0}, {0}));

return absl::OkStatus();
}

absl::Status ImageTransformer::PreInit() {
SetProcessEngine(FrameBufferUtils::ProcessEngine::kLibyuv);
return absl::OkStatus();
}

absl::Status ImageTransformer::PostInit() {
// Nothing to do.
return absl::OkStatus();
}

absl::Status ImageTransformer::CheckAndSetOutputs() {
// Nothing to do.
return absl::OkStatus();
}

StatusOr<FrameBuffer> ImageTransformer::Transform(
const FrameBuffer& frame_buffer) {
BoundingBox roi;
roi.set_width(frame_buffer.dimension().width);
roi.set_height(frame_buffer.dimension().height);
return Transform(frame_buffer, roi);
}

StatusOr<FrameBuffer> ImageTransformer::Transform(
const FrameBuffer& frame_buffer, const BoundingBox& roi) {
return InferWithFallback(frame_buffer, roi);
}

StatusOr<FrameBuffer> ImageTransformer::Postprocess(
const std::vector<const TfLiteTensor*>& /*output_tensors*/,
const FrameBuffer& /*frame_buffer*/, const BoundingBox& /*roi*/) {
ASSIGN_OR_RETURN(auto postprocessed_output, postprocessor_->Postprocess());
return postprocessed_output;
}
} // namespace vision
} // namespace task
} // namespace tflite
138 changes: 138 additions & 0 deletions tensorflow_lite_support/cc/task/vision/image_transformer.h
Original file line number Diff line number Diff line change
@@ -0,0 +1,138 @@
/* Copyright 2021 The TensorFlow Authors. All Rights Reserved.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
==============================================================================*/

#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_IMAGE_TRANSFORMER_H_
#define TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_IMAGE_TRANSFORMER_H_

#include <memory>
#include <vector>

#include "tensorflow/lite/core/api/op_resolver.h"
#include "tensorflow/lite/core/shims/cc/kernels/register.h"
#include "tensorflow_lite_support/cc/port/statusor.h"
#include "tensorflow_lite_support/cc/task/core/external_file_handler.h"
#include "tensorflow_lite_support/cc/task/vision/core/base_vision_task_api.h"
#include "tensorflow_lite_support/cc/task/vision/proto/image_transformer_options_proto_inc.h"
#include "tensorflow_lite_support/cc/task/processor/image_postprocessor.h"

namespace tflite {
namespace task {
namespace vision {

// Performs transformation on images.
//
// The API expects a TFLite model with optional, but strongly recommended,
// TFLite Model Metadata.
//
// Input tensor:
// (kTfLiteUInt8/kTfLiteFloat32)
// - image input of size `[batch x height x width x channels]`.
// - batch inference is not supported (`batch` is required to be 1).
// - only RGB inputs are supported (`channels` is required to be 3).
// - if type is kTfLiteFloat32, NormalizationOptions are required to be
// attached to the metadata for input normalization.
// At least one output tensor with:
// (kTfLiteUInt8/kTfLiteFloat32)
// - `N `classes and either 2 or 4 dimensions, i.e. `[1 x N]` or
// `[1 x 1 x 1 x N]`
// - optional (but recommended) label map(s) as AssociatedFile-s with type
// TENSOR_AXIS_LABELS, containing one label per line. The first such
// AssociatedFile (if any) is used to fill the `class_name` field of the
// results. The `display_name` field is filled from the AssociatedFile (if
// any) whose locale matches the `display_names_locale` field of the
// `ImageTransformerOptions` used at creation time ("en" by default, i.e.
// English). If none of these are available, only the `index` field of the
// results will be filled.
//
// An example of such model can be found at:
// https://tfhub.dev/bohemian-visual-recognition-alliance/lite-model/models/mushroom-identification_v1/1
//
// A CLI demo tool is available for easily trying out this API, and provides
// example usage. See:
// examples/task/vision/desktop/image_classifier_demo.cc
class ImageTransformer : public BaseVisionTaskApi<FrameBuffer> {
public:
using BaseVisionTaskApi::BaseVisionTaskApi;

// Creates an ImageTransformer from the provided options. A non-default
// OpResolver can be specified in order to support custom Ops or specify a
// subset of built-in Ops.f
static tflite::support::StatusOr<std::unique_ptr<ImageTransformer>>
CreateFromOptions(
const ImageTransformerOptions& options,
std::unique_ptr<tflite::OpResolver> resolver =
absl::make_unique<tflite_shims::ops::builtin::BuiltinOpResolver>());

// Performs actual transformation on the provided FrameBuffer.
//
// The FrameBuffer can be of any size and any of the supported formats, i.e.
// RGBA, RGB, NV12, NV21, YV12, YV21. It is automatically pre-processed before
// inference in order to (and in this order):
// - resize it (with bilinear interpolation, aspect-ratio *not* preserved) to
// the dimensions of the model input tensor,
// - convert it to the colorspace of the input tensor (i.e. RGB, which is the
// only supported colorspace for now),
// - rotate it according to its `Orientation` so that inference is performed
// on an "upright" image.
tflite::support::StatusOr<FrameBuffer> Transform(
const FrameBuffer& frame_buffer);

// Same as above, except that the transformation is performed based on the
// input region of interest. Cropping according to this region of interest is
// prepended to the pre-processing operations.
//
// IMPORTANT: as a consequence of cropping occurring first, the provided
// region of interest is expressed in the unrotated frame of reference
// coordinates system, i.e. in `[0, frame_buffer.width) x [0,
// frame_buffer.height)`, which are the dimensions of the underlying
// `frame_buffer` data before any `Orientation` flag gets applied. Also, the
// region of interest is not clamped, so this method will return a non-ok
// status if the region is out of these bounds.
tflite::support::StatusOr<FrameBuffer> Transform(
const FrameBuffer& frame_buffer, const BoundingBox& roi);

protected:
// The options used to build this ImageTransformer.
std::unique_ptr<ImageTransformerOptions> options_;

// Post-processing to transform the raw model outputs into image results.
tflite::support::StatusOr<FrameBuffer> Postprocess(
const std::vector<const TfLiteTensor*>& output_tensors,
const FrameBuffer& frame_buffer, const BoundingBox& roi) override;

// Performs sanity checks on the provided ImageTransformerOptions.
static absl::Status SanityCheckOptions(const ImageTransformerOptions& options);

// Initializes the ImageTransformer from the provided ImageTransformerOptions,
// whose ownership is transferred to this object.
absl::Status Init(std::unique_ptr<ImageTransformerOptions> options);

// Performs pre-initialization actions.
virtual absl::Status PreInit();
// Performs post-initialization actions.
virtual absl::Status PostInit();

private:
// Performs sanity checks on the model outputs and extracts their metadata.
absl::Status CheckAndSetOutputs();

std::unique_ptr<processor::ImagePostprocessor> postprocessor_;
};

} // namespace vision
} // namespace task
} // namespace tflite

#endif // TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_IMAGE_TRANSFORMER_H_
25 changes: 25 additions & 0 deletions tensorflow_lite_support/cc/task/vision/proto/BUILD
Original file line number Diff line number Diff line change
Expand Up @@ -242,3 +242,28 @@ cc_library(
hdrs = ["embeddings_proto_inc.h"],
deps = [":embeddings_cc_proto"],
)

# ImageTransformer protos.

proto_library(
name = "image_transformer_options_proto",
srcs = ["image_transformer_options.proto"],
deps = [
"//tensorflow_lite_support/cc/task/core/proto:base_options_proto",
],
)

cc_proto_library(
name = "image_transformer_options_cc_proto",
deps = [
":image_transformer_options_proto",
],
)

cc_library(
name = "image_transformer_options_proto_inc",
hdrs = ["image_transformer_options_proto_inc.h"],
deps = [
":image_transformer_options_cc_proto",
],
)
Original file line number Diff line number Diff line change
@@ -0,0 +1,28 @@
/* Copyright 2021 The TensorFlow Authors. All Rights Reserved.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
==============================================================================*/

syntax = "proto2";

package tflite.task.vision;

import "tensorflow_lite_support/cc/task/core/proto/base_options.proto";

// Options for setting up an ImageTransformer.
// Next Id: 10.
message ImageTransformerOptions {
// Base options for configuring Task library, such as specifying the TfLite
// model file with metadata, accelerator options, etc.
optional tflite.task.core.BaseOptions base_options = 1;
}
Original file line number Diff line number Diff line change
@@ -0,0 +1,23 @@
/* Copyright 2021 The TensorFlow Authors. All Rights Reserved.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
==============================================================================*/

#ifndef TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_IMAGE_TRANSFORMER_OPTIONS_PROTO_INC_H_
#define TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_IMAGE_TRANSFORMER_OPTIONS_PROTO_INC_H_

#include "tensorflow_lite_support/cc/task/core/proto/base_options_proto_inc.h"
#include "tensorflow_lite_support/cc/task/core/proto/external_file_proto_inc.h"

#include "tensorflow_lite_support/cc/task/vision/proto/image_transformer_options.pb.h"
#endif // TENSORFLOW_LITE_SUPPORT_CC_TASK_VISION_PROTO_IMAGE_TRANSFORMER_OPTIONS_PROTO_INC_H_
19 changes: 19 additions & 0 deletions tensorflow_lite_support/cc/test/task/vision/BUILD
Original file line number Diff line number Diff line change
Expand Up @@ -173,3 +173,22 @@ cc_test_with_tflite(
"@org_tensorflow//tensorflow/lite/kernels:builtin_ops",
],
)

cc_test_with_tflite(
name = "image_transformer_test",
srcs = ["image_transformer_test.cc"],
data = [
"//tensorflow_lite_support/cc/test/testdata/task/vision:test_images",
"//tensorflow_lite_support/cc/test/testdata/task/vision:test_models",
],
deps = [
"//tensorflow_lite_support/cc/task/vision:image_transformer",
"@org_tensorflow//tensorflow/lite/core/shims:cc_shims_test_util",
"//tensorflow_lite_support/cc/port:gtest_main",
"//tensorflow_lite_support/cc/task/vision/utils:frame_buffer_common_utils",
"//tensorflow_lite_support/cc/task/vision/utils:frame_buffer_utils",
"//tensorflow_lite_support/cc/test:test_utils",
"//tensorflow_lite_support/examples/task/vision/desktop/utils:image_utils",
"@com_google_absl//absl/flags:flag",
],
)
Loading