DESC:FORM 2

THE PATENTS ACT, 1970
(39 of 1970)
&
THE PATENT RULES, 2003

COMPLETE SPECIFICATION
(See Section 10 and Rule 13)

Title of invention:
METHOD AND SYSTEM FOR QUANTUM-ENHANCED IMAGE SEGMENTATION

Applicant:
Tata Consultancy Services Limited
A company Incorporated in India under the Companies Act, 1956
Having address:
Nirmal Building, 9th Floor,
Nariman Point, Mumbai 400021,
Maharashtra, India

The following specification particularly describes the invention and the manner in which it is to be performed.
CROSS-REFERENCE TO RELATED APPLICATIONS AND PRIORITY
The present application claims priority from Indian provisional patent application no. 202221012299, filed on March 07, 2022. The entire contents of the aforementioned application are incorporated herein by reference.
TECHNICAL FIELD
The disclosure herein generally relates to the field of image processing and more particularly, to a method and system for quantum-enhanced image segmentation.
BACKGROUND
Image segmentation is one of the initial steps in image analysis tasks and is being used as solution for problems such as location of objects in images, locating tumors in medical images and so on. There are various algorithms available in the art for image segmentation. Threshold based segmentation methods is one approach for image segmentation which is used largely. However, this method ignores pixel location in the image, and they result in incoherent segmentation. Another approach for image segmentation is a clustering based technique. However, the clustering based technique performs efficiently and provides good results only with a good choice of initial settings such as a number of clusters and initial cluster center locations. Also, the whole image needs to be processed in each iteration of the clustering based technique and which makes it inefficient.
Recently convolutional neural network (CNN) models have also been used for image segmentation. However, these methods are very long learning process, and they need a huge and diverse dataset for training the model. If the model is not trained properly, the results for the image segmentation will be poor. Image segmentation is a hard task in image processing when the images are highly complex. For example, images which comprise very fine cracks in the case of images of metallic parts wherein the cracks are minute when compared to the whole image.
Nowadays, classical U-Nets are used for image segmentation. Classical U-Net is a U-shaped topology where the images are repeatedly passed through convolutional filters and pooling layers to reduce their size while increasing the number of channels. When a bottleneck layer (the middle layers of U-Net) is reached, the tensors are then up convolved to increase their size back to an extent and gradually reduce the number of channels to match the number of classes of objects that need to be segmented. The tensors from the downscaling layers are concatenated with corresponding ones of the upscaling layers along the way. Even though the classical U-Nets perform accurately, training time for the classical U-Nets are high.
SUMMARY
Embodiments of the present disclosure present technological improvements as solutions to one or more of the above-mentioned technical problems recognized by the inventors in conventional systems. For example, in one embodiment, a method for quantum-enhanced image segmentation is provided. The method includes: receiving (i) a set of training images and (ii) a set of annotated images corresponding to the set of training images as training input for training a quantum-classical hybrid model for image segmentation, wherein the quantum-classical hybrid model includes a U-Net architecture comprising a bottleneck layer including (i) a classical path and (ii) a quantum path; training the quantum-classical hybrid model using the training input until a training loss associated with the quantum-classical hybrid model converges to a predefined value; receiving an input image for segmenting; preprocessing the input image using a set of preprocessing techniques; encoding the input image using a set of convolutional and pooling layers present in an encoder in the trained quantum-classical hybrid model; obtaining from the encoded input image (i) a first tensor from the classical path and (ii) a second tensor from the quantum path of the bottleneck layer of the trained quantum-classical hybrid model; concatenating the first tensor and the second tensor to obtain a merged tensor corresponding to the input image; and obtaining a segmented image by decoding the merged tensor corresponding to the input image.
In another aspect, a system for quantum-enhanced image segmentation is provided. The system comprises memory storing instructions; one or more communication interfaces; and one or more hardware processors coupled to the memory via the one or more communication interfaces, wherein the one or more hardware processors are configured by the instructions to receive (i) a set of training images and (ii) a set of annotated images corresponding to the set of training images as training input for training a quantum-classical hybrid model for image segmentation, wherein the quantum-classical hybrid model includes a U-Net architecture comprising a bottleneck layer including (i) a classical path and (ii) a quantum path; train the quantum-classical hybrid model using the training input until a training loss associated with the quantum-classical hybrid model converges to a predefined value; receive an input image for segmenting; preprocessing the input image using a set of preprocessing techniques; encode the input image using a set of convolutional and pooling layers present in an encoder in the trained quantum-classical hybrid model; obtain from the encoded input image (i) a first tensor from the classical path and (ii) a second tensor from the quantum path of the bottleneck layer of the trained quantum-classical hybrid model; concatenate the first tensor and the second tensor to obtain a merged tensor corresponding to the input image; and obtain a segmented image by decoding the merged tensor corresponding to the input image.
In an embodiment, wherein the training of the quantum-classical hybrid model comprises, preprocessing the set of training images using the set of preprocessing techniques; encoding the set of training images using the set of convolutional and pooling layers present in the encoder of the quantum-classical hybrid model; obtaining from each of the training image of the set of training images (i) a first tensor from the classical path and (ii) a second tensor from the quantum path of the bottleneck layer of the quantum-classical hybrid model; concatenating the first tensor and the second tensor to obtain a merged tensor corresponding to each of the training image; obtaining a set of segmented images by decoding the merged tensor using a set of convolutional and pooling layers present in a decoder of the quantum-classical hybrid model; and calculating the training loss based on (i) the set of segmented images and (ii) the set of annotated images.
In an embodiment, the training of the quantum-classical hybrid model further comprises, backpropagating the training loss to the encoder through the classical path of the bottleneck layer of the quantum-classical hybrid model; and tuning a set of parameters in the quantum-classical hybrid model based on the training loss.
In an embodiment, wherein the encoding of set of training images reduces dimensions of the set of training images.
In an embodiment, wherein obtaining the second tensor from each training image further comprises, obtaining the second tensor by mapping the encoded training image using a quantum feature map comprising, passing a set of patches of the encoded training image to the quantum path; flattening each patch of the set of patches into an array comprising a set of elements; mapping the set of elements in the array to a set of qubits using a quantum circuit corresponding to the quantum feature map; obtaining a set of expectation values by calculating an expectation value associated with each qubit of the set of qubits; mapping the set of expectation values to a same channel to obtain the second tensor; convolving the second tensor to a dimension of the first tensor using a down convolution layer present in the quantum path of the bottleneck layer.
In an embodiment, wherein the quantum feature map is anyone of (i) parameterized or (ii) non-parameterized.
In an embodiment, providing a unidirectional cut to a layer following the quantum path if the quantum feature map is non-parameterized.
In an embodiment, wherein the set of training images and the input image are real world images of resolution having anyone of (i) greater than 2048x1080 pixels or (ii) smaller than 2048x1080 pixels.
In yet another aspect, there is provided a computer program product comprising a non-transitory computer readable medium having a computer readable program embodied therein, wherein the computer readable program, when executed on a computing device causes the computing device for quantum-enhanced image segmentation by receiving (i) a set of training images and (ii) a set of annotated images corresponding to the set of training images as training input for training a quantum-classical hybrid model for image segmentation, wherein the quantum-classical hybrid model includes a U-Net architecture comprising a bottleneck layer including (i) a classical path and (ii) a quantum path; training the quantum-classical hybrid model using the training input until a training loss associated with the quantum-classical hybrid model converges to a predefined value; receiving an input image for segmenting; preprocessing the input image using a set of preprocessing techniques; encoding the input image using a set of convolutional and pooling layers present in an encoder in the trained quantum-classical hybrid model; obtaining from the encoded input image (i) a first tensor from the classical path and (ii) a second tensor from the quantum path of the bottleneck layer of the trained quantum-classical hybrid model; concatenating the first tensor and the second tensor to obtain a merged tensor corresponding to the input image; and obtaining a segmented image by decoding the merged tensor corresponding to the input image.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed.

BRIEF DESCRIPTION OF THE DRAWINGS
The accompanying drawings, which are incorporated in and constitute a part of this disclosure, illustrate exemplary embodiments and, together with the description, serve to explain the disclosed principles:
FIG. 1 illustrates an exemplary block diagram of a system for quantum-enhanced image segmentation, in accordance with some embodiments of the present disclosure.
FIG. 2 is an exemplary flow diagram for a method for quantum-enhanced image segmentation using a trained quantum-classical hybrid image segmentation model according to some embodiments of the present disclosure.
FIG. 3 is an exemplary flow diagram depicting steps for training the quantum-classical hybrid image segmentation model for image segmentation according to some embodiments of the present disclosure
FIG. 4 depicts an example U-Net architecture for image segmentation according to some embodiments of the present disclosure.
FIG. 5 depicts an example bottleneck layer of compressed U-Net architecture for image segmentation according to some embodiments of the present disclosure.
FIG. 6 depicts an example bottleneck layer of the quantum-classical hybrid image segmentation model for image segmentation according to some embodiments of the present disclosure.
FIG. 7 depicts an example circuit pertaining to a single layer of ZZ feature map for image segmentation according to some embodiments of the present disclosure.
FIG. 8A and FIG. 8B depicts segmentation result images using a compressed U-Net and the quantum-classical hybrid image segmentation model respectively according to some embodiments of the present disclosure.
FIG.9 is a graphical illustration showing value of Binary Cross-entropy Loss at each epoch while training the compressed U-Net and the quantum-classical hybrid image segmentation model according to some embodiments of the present disclosure.
DETAILED DESCRIPTION OF EMBODIMENTS
Exemplary embodiments are described with reference to the accompanying drawings. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. Wherever convenient, the same reference numbers are used throughout the drawings to refer to the same or like parts. While examples and features of disclosed principles are described herein, modifications, adaptations, and other implementations are possible without departing from the scope of the disclosed embodiments.
The embodiments herein provide a method and system for quantum-enhanced image segmentation. The embodiments disclose a quantum-classical hybrid image segmentation model for image segmentation which enhances the performance of a classical U-Net architecture by interleaving quantum layers with classical ones. The quantum-classical hybrid image segmentation model comprises quantum layers and layers of classical U-Net architecture which are trained together for performing image segmentation. The example model uses quantum convolution at bottleneck layers of the U-Net and in future with the larger and quality quantum computers become available, the quantum layer can be utilized in other layers of classical U-Net also, not restricting to bottleneck layer alone.
The quantum-classical hybrid nature of the invention is used for segmenting large size real-life images of practical interest. The disclosed model is configurable depending on the capability of the available quantum computer at that time.
Conventionally, in quantum U-Net architectures for image classification, features are extracted from pretrained deep learning models from images. These features are then transformed to another feature space by embedding them into a quantum circuit and measuring in an appropriate basis. The quantum transformed features are then either directly used to draw inference or are fed into a multilayered perceptron for single/multi-class classification.
Referring now to the drawings, and more particularly to FIG. 1 through FIG.-9, where similar reference characters denote corresponding features consistently throughout the figures, there are shown preferred embodiments and these embodiments are described in the context of the following exemplary system and/or method.
FIG. 1 illustrates a system 100 for quantum-enhanced image segmentation. In an embodiment, the system 100 includes one or more processors 102, communication interface device(s) or input/output (I/O) interface(s) 106, and one or more data storage devices or memory 104 operatively coupled to the one or more processors 102. The one or more processors 102 that are hardware processors can be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, graphics controllers, logic circuitries, and/or any devices that manipulate signals based on operational instructions. Among other capabilities, the processor(s) are configured to fetch and execute computer-readable instructions stored in the memory. In the context of the present disclosure, the expressions ‘processors’ and ‘hardware processors’ may be used interchangeably. In an embodiment, the system 100 can be implemented in a variety of computing systems, such as laptop computers, notebooks, hand-held devices, workstations, mainframe computers, servers, a network cloud and the like.
The I/O interface (s) 106 may include a variety of software and hardware interfaces, for example, a web interface, a graphical user interface, and the like and can facilitate multiple communications within a wide variety of networks and protocol types, including wired networks, for example, LAN, cable, etc., and wireless networks, such as WLAN, cellular, or satellite. In an embodiment, the I/O interface(s) can include one or more ports for connecting a number of devices to one another or to another server.
The memory 104 may include any computer-readable medium known in the art including, for example, volatile memory, such as static random access memory (SRAM) and dynamic random access memory (DRAM), and/or non-volatile memory, such as read only memory (ROM), erasable programmable ROM, flash memories, hard disks, optical disks, and magnetic tapes. The memory 104 may also include quantum random access memory (qRAM) which uses qubits to address any quantum superposition of memory cells. The quantum modules explained further can be implemented on quantum processors based on various technologies such as superconducting qubits, ion trapped qubits and so on.
FIG. 2 is an exemplary flow diagram for a method for quantum-enhanced image segmentation using a trained quantum-classical hybrid image segmentation model according to some embodiments of the present disclosure.
In an embodiment of the present disclosure, the one or more processors 102 are configured to receive at step 202, (i) a set of training images and (ii) a set of annotated images corresponding to the set of training images as training input for training a quantum-classical hybrid model for image segmentation. The quantum-classical hybrid model includes a U-Net architecture comprising a bottleneck layer including (i) a classical path and (ii) a quantum path.
In an embodiment of the present disclosure, the one or more processors 102 are configured to train at step 204, the quantum-classical hybrid model using the training input until a training loss associated with the quantum-classical hybrid model converges to a predefined value. FIG.3 is an exemplary flow diagram depicting the steps for training the quantum-classical hybrid image segmentation model for image segmentation according to some embodiments of the present disclosure.
The training process of the quantum-classical hybrid image segmentation model is explained hereafter. In an embodiment of the present disclosure, the one or more processors 102 are configured to preprocess at step 302 the set of training images using the set of preprocessing techniques. The preprocessing techniques includes rescaling, translation and so on which are performed using conventional methods. In an embodiment of the present disclosure, the one or more processors 102 are configured to encode at step 304 the preprocessed set of training images using a set of convolutional and pooling layers present in the encoder of the quantum-classical hybrid model. The training images are encoded through a sequence of convolution and pooling layers to gradually reduce its dimensions with a corresponding increase in the number of channels.
Further in the training process, in an embodiment of the present disclosure, the one or more processors 102 are configured to obtain at step 306 (i) a first tensor from the classical path and (ii) a second tensor from the quantum path of the bottleneck layer of the quantum-classical hybrid model from each of the training image of the set of training images. The bottleneck layer of the quantum-classical hybrid image segmentation model bifurcates into the classical path and the quantum path. Along the classical path, there is a convolutional layer with the first tensor as output with a size of m×n×p. Along the quantum path, w×w patches of the encoded image are passed to a quantum feature-map (the present disclosure has considered single-layered, non-parametrized ZZ feature map; however, a parameterized feature map can be considered as well), one channel at a time. The resultant wm×wn×p tensor obtained by measuring the quantum circuits are brought down to the second tensor with a size of m×n×p using down convolution to match the shape of the first tensor in the classical path.
In an embodiment, the second tensor is obtained by mapping the encoded input image using a quantum feature map. The process of obtaining the second tensor is explained hereafter. A set of patches (each patch of size (w × w))of the encoded input image are passed to the quantum path. Further each patch is flattened into an array of size w^2. The array comprises a set of elements. The set of elements is mapped to a set of qubits using a quantum circuit corresponding to the quantum feature map. Further a set of expectation values is obtained by calculating an expectation value associated with each qubit of the set of qubits. These set of expectation values are mapped to a same channel to obtain the second tensor. After obtaining the second tensor, it is convolved to a dimension of the first tensor using a down convolution layer present in the quantum path of the bottleneck layer.
The disclosed model introduces the concept of parallel paths, or subnetworks with unidirectional propagation. The encoded image is acted upon by classical convolution to get a m×n×p first tensor. Simultaneously, the same encoded image is passed as an input to the quantum path. When the quantum feature-map is non-parametrized, a unidirectional cut is made to the network following the quantum layer that produces the wm×wn×p tensor as an output. The cut is such that computations happen freely along the quantum path during forward propagation, but back-propagation through the quantum path is restricted. As already mentioned, the wm×wn×p tensor is reduced to m×n×p, the result of which is concatenated with the m×n×p output from the classical path to get a merged m×n×2p tensor along the channel axis.
In the state-of-the art compressed U-Net model the desired quantum advantage is achieved using quantum convolution at the bottleneck layers of U-Net. FIG.4 depicts an example U-Net architecture for image segmentation according to some embodiments of the present disclosure. FIG.5 depicts an example bottleneck layer of the compressed U-Net architecture for image segmentation according to some embodiments of the present disclosure. The tensors in the bottleneck layer are mapped to another feature space using a quantum feature map. The classically-intractable mapping of features translates them to a space which may accord some benefit over the original space to the classification process. Ideally, the circuit for feature embedding and translation can be learnt by utilizing optimizable parameters.
FIG.6 depicts an example of the bottleneck layer of the quantum-classical hybrid image segmentation model for image segmentation according to some embodiments of the present disclosure. As mentioned earlier, parallel quantum path (subnetwork with unidirectional propagation) is introduced in the bottleneck layer, where w × w patches are extracted channel-wise and convolved by a quantum circuit consisting of the quantum feature-map and w^2 qubits.
In the FIG.6, a 32x32x8 tensor is passed as an input to the quantum convolutional procedure. A unidirectional cut is made to the network following the quantum layer that produces the 90x90x8 tensor as an output. The cut is such that computations happen freely along the quantum path during forward propagation, but back-propagation through the quantum layer is restricted. As already mentioned, the 90x90x8 tensor is reduced to 30x30x8, the result of which is concatenated with the 30x30x8 output from the classical path to get a resultant 30x30x16 tensor.
FIG.7 depicts an example circuit pertaining to a single layer of the ZZ feature map for image segmentation according to some embodiments of the present disclosure. All nine qubits are measured at the end of the circuit and their expectation values in the Z-basis are mapped into a w × w matrix. The process is repeated for all the w × w patches across all the channels to finally obtain the wm×wn×p tensor, which is then brought down to m×n×p through classical convolution with a w × w kernel and a stride of w. The ZZ feature map is utilized to embed and transform the input features to the quantum circuit. Although it is usually used as a kernel function in Quantum Support-Vector Machines, they can also serve as feature maps due to the equivalence of kernel functions and feature maps. The advantages of using the ZZ feature maps lie in the fact that they are non-parametrized and have been demonstrated to provide advantages over classically possible techniques. The feature map was restricted to having only a single layer during experimentation to keep the simulation times in check.
As shown in FIG.7 the w × w input patch is flattened, and each of the elements is mapped to a quantum bit or qubit. The embedding is done via the use of H, RZ and RZZ gates. The xi represent the input element values that are appropriately scaled to lie in the range of [0,p]. ZZ coupling is used to capture the interaction of the central element (mapped to the 4th qubit when counting starts from 0) with its neighbors, i.e.,?_i=px_4 x_i.
Fig.7 is only exemplary and the circuit or the quantum feature-map can take variety of forms including ones with tunable parameters.
In an embodiment of the present disclosure, the one or more processors 102 are configured to concatenate at step 308 the first tensor and the second tensor to obtain a merged tensor corresponding to each of the training image. The tensors from the classical and quantum paths are concatenated along the channel axis to get the merged tensor of size m×n×2p tensor. The merged tensor is then further down convolved and pooled in the rest of the bottleneck layer.
In an embodiment of the present disclosure, the one or more processors 102 are configured to obtain at step 310 a set of segmented images by decoding the merged tensor using a set of convolutional and pooling layers present in a decoder of the quantum-classical hybrid model. The output of the bottleneck layer is decoded using up and down convolutions and pooling.
In an embodiment of the present disclosure, the one or more processors 102 are configured to calculate at step 312 the training loss based on (i) the set of segmented images and (ii) the set of annotated images. The training of the quantum-classical hybrid model further comprises, backpropagating the training loss to the encoder through the classical path of the bottleneck layer of the quantum-classical hybrid model and further tuning a set of parameters in the quantum-classical hybrid model based on the training loss. Based on the loss value, backward propagation takes place normally along the decoder, until the bottleneck layer. At the bottleneck layer, the backpropagation passes unhindered along the classical path, but stops at quantum module on the quantum path. Backpropagation continues in the encoder through the classical path. The steps 302 through 312 are repeated for multiple epochs for training the quantum-classical hybrid image segmentation model until the training loss is minimized and it converges to a predefined value, tuning the parameters in the model.
In an embodiment of the present disclosure, the one or more processors 102 are configured to receive at step 206 an input image for segmenting. The training images and the input image considered in the disclosed method may be real world images of resolution greater than 2048x1080 pixels (2K). However, it can also be configured with images of resolution lesser than 2K.
Further in an embodiment of the present disclosure, the one or more processors 102 are configured to preprocess the input image using the set of preprocessing techniques. The same set of preprocessing techniques as used in the training process of the quantum-classical hybrid image segmentation model is being used.
In an embodiment of the present disclosure, the one or more processors 102 are configured to encode at step 210 the input image using the set of convolutional and pooling layers present in an encoder in the trained quantum-classical hybrid model. In an embodiment of the present disclosure, at step 212 the one or more processors 102 are configured to obtain (i) a first tensor from the classical path and (ii) a second tensor from the quantum path of the bottleneck layer of the trained quantum-classical hybrid model from the encoded input image. The second tensor is obtained by mapping the encoded input image using a quantum feature map. The process explained for step 306 is being utilized for obtaining the second tensor from the encoded input image.
In an embodiment of the present disclosure, at step 212 the one or more processors 102 are configured to concatenate the first tensor and the second tensor to obtain a merged tensor corresponding to the input image.
Further, in an embodiment of the present disclosure, at step 212 the one or more processors 102 are configured to obtain a segmented image by decoding the merged tensor corresponding to the input image.
EXPERIMENTAL RESULTS: The performance of the disclosed model and the Compressed U-Net (CU-Net) were evaluated simultaneously. Both the models were trained on 50 images and validated on 20 images from the Kaggle Surface Crack Detection dataset. The crack segmentation results of the CU-Net and the disclosed model are shown in FIG.8A and FIG.8B respectively. FIG.8A and FIG.8B depicts segmentation result images using a compressed U-Net and the quantum-classical hybrid image segmentation model respectively according to some embodiments of the present disclosure. Also, experiments were performed using real dataset. It was observed that the training and validation losses were reduced rapidly on the disclosed model for both the real dataset and Kaggle dataset. The background pixels get faded sooner through quantum, which explain the quicker reduction of loss. Rate of loss-reduction and fading of background through the disclosed quantum model was found to be higher than CU-Net. Table 1 shows the Intersection over Unio (IoU) scores over the validation set images for CU-Net and the disclosed model at various epochs for the real datasets and Kaggle dataset. FIG.9 is a graphical illustration showing value of Binary Cross-entropy Loss at each epoch while training the compressed U-Net and the quantum-classical hybrid image segmentation model according to some embodiments of the present disclosure. The graph indicates that the loss reduced with a pace that is more rapid for the disclosed model, compared to CU-Net. The observation was corroborated with the Kaggle dataset as well, where the difference was even higher and more marked.

Dataset Epoch Number CU-Net Disclosed model
Real dataset1 55 36.18 45.41
Real dataset2 70 48.93 50.16
Kaggle dataset 110 69.64 73.17
Table 1
The written description describes the subject matter herein to enable any person skilled in the art to make and use the embodiments. The scope of the subject matter embodiments is defined by the claims and may include other modifications that occur to those skilled in the art. Such other modifications are intended to be within the scope of the claims if they have similar elements that do not differ from the literal language of the claims or if they include equivalent elements with insubstantial differences from the literal language of the claims.
The embodiments of present disclosure herein address the problem of image segmentation using quantum-classical hybrid model. The embodiment thus provides the method steps for performing image segmentation using the quantum- classical hybrid model. The disclosed quantum-classical hybrid model includes U-Net architecture comprising a bottleneck layer of classical path and quantum path. Moreover, the quantum-classical hybrid model comprises quantum layers and layers of classical U-Net architecture which are trained together for performing image segmentation. The disclosed method is used for segmenting very large sized real life images.
It is to be understood that the scope of the protection is extended to such a program and in addition to a computer-readable means having a message therein; such computer-readable storage means contain program-code means for implementation of one or more steps of the method, when the program runs on a server or mobile device or any suitable programmable device. The hardware device can be any kind of device which can be programmed including e.g., any kind of computer like a server or a personal computer, or the like, or any combination thereof. The device may also include means which could be e.g., hardware means like e.g., an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination of hardware and software means, e.g., an ASIC and an FPGA, or at least one microprocessor and at least one memory with software processing components located therein. Thus, the means can include both hardware means and software means. The method embodiments described herein could be implemented in hardware and software. The device may also include software means. Alternatively, the embodiments may be implemented on different hardware devices, e.g., using a plurality of CPUs.
The embodiments herein can comprise hardware and software elements. The embodiments that are implemented in software include but are not limited to, firmware, resident software, microcode, etc. The functions performed by various components described herein may be implemented in other components or combinations of other components. For the purposes of this description, a computer-usable or computer readable medium can be any apparatus that can comprise, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
The illustrated steps are set out to explain the exemplary embodiments shown, and it should be anticipated that ongoing technological development will change the manner in which particular functions are performed. These examples are presented herein for purposes of illustration, and not limitation. Further, the boundaries of the functional building blocks have been arbitrarily defined herein for the convenience of the description. Alternative boundaries can be defined so long as the specified functions and relationships thereof are appropriately performed. Alternatives (including equivalents, extensions, variations, deviations, etc., of those described herein) will be apparent to persons skilled in the relevant art(s) based on the teachings contained herein. Such alternatives fall within the scope of the disclosed embodiments. Also, the words “comprising,” “having,” “containing,” and “including,” and other similar forms are intended to be equivalent in meaning and be open ended in that an item or items following any one of these words is not meant to be an exhaustive listing of such item or items, or meant to be limited to only the listed item or items. It must also be noted that as used herein and in the appended claims, the singular forms “a,” “an,” and “the” include plural references unless the context clearly dictates otherwise.
Furthermore, one or more computer-readable storage media may be utilized in implementing embodiments consistent with the present disclosure. A computer-readable storage medium refers to any type of physical memory on which information or data readable by a processor may be stored. Thus, a computer-readable storage medium may store instructions for execution by one or more processors, including instructions for causing the processor(s) to perform steps or stages consistent with the embodiments described herein. The term “computer-readable medium” should be understood to include tangible items and exclude carrier waves and transient signals, i.e., be non-transitory. Examples include random access memory (RAM), read-only memory (ROM), volatile memory, nonvolatile memory, hard drives, CD ROMs, DVDs, flash drives, disks, and any other known physical storage media.
It is intended that the disclosure and examples be considered as exemplary only, with a true scope of disclosed embodiments being indicated by the following claims.
,CLAIMS:
1. A processor implemented method (200) comprising:
receiving (202), via one or more hardware processors, (i) a set of training images and (ii) a set of annotated images corresponding to the set of training images as training input for training a quantum-classical hybrid model for image segmentation, wherein the quantum-classical hybrid model includes a U-Net architecture comprising a bottleneck layer including (i) a classical path and (ii) a quantum path;
training (204) the quantum-classical hybrid model, via the one or more hardware processors, using the training input until a training loss associated with the quantum-classical hybrid model converges to a predefined value;
receiving (206), via the one or more hardware processors, an input image for segmenting;
preprocessing (208), via the one or more hardware processors, the input image using a set of preprocessing techniques;
encoding the input image (210), via the one or more hardware processors, using a set of convolutional and pooling layers present in an encoder in the trained quantum-classical hybrid model;
obtaining (212), via the one or more hardware processors, from the encoded input image (i) a first tensor from the classical path and (ii) a second tensor from the quantum path of the bottleneck layer of the trained quantum-classical hybrid model;
concatenating (214), via the one or more hardware processors, the first tensor and the second tensor to obtain a merged tensor corresponding to the input image; and
obtaining (216), via the one or more hardware processors, a segmented image by decoding the merged tensor corresponding to the input image.

2. The method of claim 1, wherein the training of the quantum-classical hybrid model (300) comprises,
preprocessing (302), via the one or more hardware processors, the set of training images using the set of preprocessing techniques;
encoding (304) the set of training images, via the one or more hardware processors, using the set of convolutional and pooling layers present in the encoder of the quantum-classical hybrid model;
obtaining (306), via the one or more hardware processors, from each of the training image of the set of training images (i) a first tensor from the classical path and (ii) a second tensor from the quantum path of the bottleneck layer of the quantum-classical hybrid model;
concatenating (308), via the one or more hardware processors, the first tensor and the second tensor to obtain a merged tensor corresponding to each of the training image;
obtaining (310) a set of segmented images, via the one or more hardware processors, by decoding the merged tensor using a set of convolutional and pooling layers present in a decoder of the quantum-classical hybrid model; and
calculating (312), via the one or more hardware processors, the training loss based on (i) the set of segmented images and (ii) the set of annotated images.

3. The method of claim 2, the training of the quantum-classical hybrid model further comprises,
backpropagating, via the one or more hardware processors, the training loss to the encoder through the classical path of the bottleneck layer of the quantum-classical hybrid model; and
tuning, via the one or more hardware processors, a set of parameters in the quantum-classical hybrid model based on the training loss.

4. The method of claim 2, wherein the encoding of set of training images reduces dimensions of the set of training images

5. The method of claim 2, wherein obtaining the second tensor from each training image further comprises,
obtaining the second tensor by mapping the encoded training image using a quantum feature map comprising,
passing a set of patches of the encoded training image to the quantum path;
flattening each patch of the set of patches into an array comprising a set of elements;
mapping the set of elements in the array to a set of qubits using a quantum circuit corresponding to the quantum feature map;
obtaining a set of expectation values by calculating an expectation value associated with each qubit of the set of qubits;
mapping the set of expectation values to a same channel to obtain the second tensor;
convolving the second tensor to a dimension of the first tensor using a down convolution layer present in the quantum path of the bottleneck layer.

6. The method of claim 5, wherein the quantum feature map is anyone of (i) parameterized or (ii) non-parameterized.

7. The method of claim 5, further comprises providing a unidirectional cut to a layer following the quantum path if the quantum feature map is non-parameterized.

8. The method of claim 1, wherein the set of training images and the input image are real world images of resolution having anyone of (i) greater than 2048x1080 pixels or (ii) smaller than 2048x1080 pixels.

9. A system (100), comprising:
a memory (104) storing instructions;
one or more communication interfaces (106); and
one or more hardware processors (102) coupled to the memory (102) via the one or more communication interfaces (106), wherein the one or more hardware processors (102) are configured by the instructions to:
receive (i) a set of training images and (ii) a set of annotated images corresponding to the set of training images as training input for training a quantum-classical hybrid model for image segmentation, wherein the quantum-classical hybrid model includes a U-Net architecture comprising a bottleneck layer including (i) a classical path and (ii) a quantum path;
train the quantum-classical hybrid model using the training input until a training loss associated with the quantum-classical hybrid model converges to a predefined value;
receive an input image for segmenting;
preprocess the input image using a set of preprocessing techniques;
encode the input image using a set of convolutional and pooling layers present in an encoder in the trained quantum-classical hybrid model;
obtain from the encoded input image (i) a first tensor from the classical path and (ii) a second tensor from the quantum path of the bottleneck layer of the trained quantum-classical hybrid model;
concatenate the first tensor and the second tensor to obtain a merged tensor corresponding to the input image; and
obtain a segmented image by decoding the merged tensor corresponding to the input image.

10. The system of claim 9, wherein the training of the quantum-classical hybrid model comprises,
preprocess the set of training images using the set of preprocessing techniques;
encode the set of training images using the set of convolutional and pooling layers present in the encoder of the quantum-classical hybrid model;
obtain from each of the training image of the set of training images (i) a first tensor from the classical path and (ii) a second tensor from the quantum path of the bottleneck layer of the quantum-classical hybrid model;
concatenate the first tensor and the second tensor to obtain a merged tensor corresponding to each of the training image;
obtain a set of segmented images by decoding the merged tensor using a set of convolutional and pooling layers present in a decoder of the quantum-classical hybrid model; and
calculate the training loss based on (i) the set of segmented images and (ii) the set of annotated images.

11. The system of claim 10, the training of the quantum-classical hybrid model further comprises,
backpropagate the training loss to the encoder through the classical path of the bottleneck layer of the quantum-classical hybrid model; and
tune a set of parameters in the quantum-classical hybrid model based on the training loss.

12. The system of claim 10, wherein the encoding of set of training images reduces dimensions of the set of training images.

13. The system of claim 10, wherein obtaining the second tensor from each training image further comprises,
obtaining the second tensor by mapping the encoded training image using a quantum feature map comprising,
passing a set of patches of the encoded training image to the quantum path;
flattening each patch of the set of patches into an array comprising a set of elements;
mapping the set of elements in the array to a set of qubits using a quantum circuit corresponding to the quantum feature map;
obtaining a set of expectation values by calculating an expectation value associated with each qubit of the set of qubits;
mapping the set of expectation values to a same channel to obtain the second tensor;
convolving the second tensor to a dimension of the first tensor using a down convolution layer present in the quantum path of the bottleneck layer.

14. The system of claim 13, wherein the quantum feature map is anyone of (i) parameterized or (ii) non-parameterized.

15. The system of claim 13, further comprises providing a unidirectional cut to a layer following the quantum path if the quantum feature map is non-parameterized.

16. The system of claim 9, wherein the set of training images and the input image are real world images of resolution having anyone of (i) greater than 2048x1080 pixels or (ii) smaller than 2048x1080 pixels.