Jadhav, Rohan Indrajeet · TRAP@NCI (National College of Ireland) · 2022
Categorizing of clothes in the wild fashion images involves identifying the type of clothes a person wears from non-studio images such as a shirt, trousers, and so on. Identifying the fashion clothes from wild images that are grainy, unfocused, with people in different poses is a challenge. This research proposes the instance segmentation model outperforms compare to object detection model when it comes to clothes categorization from wild images. Faster RCNN is and object detection deep learning model and the segmentation model used is Mask RCNN which is based on RCNN (Region-Based Convolutional Neural Network). Both models are trained on the subset of DeepFashion2 dataset. The models are trained on 10k distinct fashion images using rich annotation files in file, 2k images are used for test purposes and 5k images for the validation. The results of two models presented in this paper based on the average precision, recall across the different IoU(Intersection over Union) metrics considering the clothes acquired in fashion image. Based on the results it is shown as the segmentation model Mask, the RCNN outperforms by 20% compared to object detection model Faster RCNN.
Cited by 0
Mehak Sheikh, Muhammad Adeel Asghar, Ruqia Bibi et al. · Sensors · 2022
Unlike 2-dimensional (2D) images, direct 3-dimensional (3D) point cloud processing using deep neural network architectures is challenging, mainly due to the lack of explicit neighbor relationships. Many researchers attempt to remedy this by performing an additional voxelization preprocessing step. However, this adds additional computational overhead and introduces quantization error issues, limiting an accurate estimate of the underlying structure of objects that appear in the scene. To this end, in this article, we propose a deep network that can directly consume raw unstructured point clouds to perform object classification and part segmentation. In particular, a Deep Feature Transformation Network (DFT-Net) has been proposed, consisting of a cascading combination of edge convolutions and a feature transformation layer that captures the local geometric features by preserving neighborhood relationships among the points. The proposed network builds a graph in which the edges are dynamically and independently calculated on each layer. To achieve object classification and part segmentation, we ensure point order invariance while conducting network training simultaneously-the evaluation of the proposed network has been carried out on two standard benchmark datasets for object classification and part segmentation. The results were comparable to or better than existing state-of-the-art methodologies. The overall score obtained using the proposed DFT-Net is significantly improved compared to the state-of-the-art methods with the ModelNet40 dataset for object categorization.
Cited by 7Open access
Chih‐Fong Tsai, Ya‐Han Hu, Wei‐Chao Lin · 2016
The success of object categorization is heavily dependent on the extracted image descriptors. In general, image or region segmentation is usually performed to segment an image into several regions or objects, and then some level-level features, such as color and texture, are extracted from each region. As a result, the region descriptor or the combination of multiple region descriptors can be used to represent a specific object or the whole image for categorization. Since there are many well-known region segmentation algorithms proposed in literature, and using different region segmentation algorithms can produce different region descriptors for the same images, no study examines the effect of region segmentation on object categorization. In this paper, we apply three well-known region segmentation algorithms for image feature extraction and representation, which are graph cuts, mean-shift segmentation, and normalized cuts. Then, the support vector machine (SVM) is used as the classifier for object categorization. Our experimental results based on Caltech 5, Caltech 8, and Corel 10 datasets show that the normalized cuts algorithm performs best. In addition, the image feature representation based on multiple region descriptors can provide more discriminative power than using center region descriptors.
Cited by 1
Riccardo Monica, Jacopo Aleotti · IEEE Robotics and Automation Letters · 2020
This work presents an approach for part-based grasp planning in point clouds. A complete pipeline is proposed that allows a robot manipulator equipped with a range camera to perform object detection, categorization, segmentation into meaningful parts, and part-based semantic grasping. A supervised image-space technique is adopted for point cloud segmentation based on projective analysis. Projective analysis generates a set of 2D projections from the input object point cloud, labels each object projection by transferring knowledge from existing labeled images, and then fuses the labels by back-projection on the object point cloud. We introduce an algorithm for point cloud categorization based on 2D projections. We also propose a viewpoint aware algorithm that filters 2D projections according to the scanning path of the robot. Object categorization and segmentation experiments were carried out with both synthetic and real datasets. Results indicate that the proposed approach performs better than a CNN-based method for a training set of limited size. Finally, we show part-based grasping tasks in a real robotic setup.
Cited by 21
Yi-hsuan Tsai, Jimei Yang, Ming-hsuan Yang · 2013
We present a learning algorithm for joint object segmentation and categorization that decomposes the original problem into two sub-tasks and admits their bidirectional interaction. In the first stage, in order to decompose output space, we train category-specific segmentation models to generate figure-ground hypotheses. In the second stage, by taking advantage of object figure-ground information, we train a multi-class segment-based categorization model to determine the object class. A re-ranking strategy is then applied to classified segments to obtain the final category-level segmentation results. Experiments on the Graz-02 and Caltech-101 datasets show that the proposed algorithm performs favorably against the state-of-the-art methods. 1
Cited by 0
Iasonas Kokkinos, Petros Maragos · DSpace - NTUA (National Technical University of Athens) · 2005
In this work, we deal with the problem of modelling and exploiting the interaction between the processes of image segmentation and object categorization. We propose a novel framework to address this problem that is based on the combination of the expectation maximization (EM) algorithm and generative models for object categories. Using a concise formulation of the interaction between these two processes, segmentation is interpreted as the E step, assigning observations to models, whereas object detection/analysis is modelled as the M-step, fitting models to observations. We present in detail the segmentation and detection processes comprising the E and M steps and demonstrate results on the joint detection and segmentation of the object categories of faces and cars.
Cited by 15Open access
Iasonas Kokkinos, Petros Maragos · 2005
In this work we deal with the problem of modelling and exploiting the interaction between the processes of image segmentation and object categorization. We propose a novel framework to address this problem that is based on the combination of the Expectation Maximization (EM) algorithm and generative models for object categories. Using a concise formulation of the interaction between these two processes, segmentation is interpreted as the E step, assigning observations to models, whereas object detection/analysis is modelled as the M-step, fitting models to observations. We present in detail the segmentation and detection processes comprising the E and M steps and demonstrate results on the joint detection and segmentation of the object categories of faces and cars.
Cited by 0
Chenhao Xu, Chang‐Tsun Li, Yongjian Hu et al. · Pattern Recognition · 2025
Video Instance Segmentation (VIS), also known as multi-object tracking and segmentation, represents a fundamental challenge in computer vision that requires simultaneous detection, segmentation, and tracking of object instances across video frames. This complex task has gained significant attention due to its crucial role in various real-world applications. The advent of deep learning has promoted VIS approaches, leading to numerous architectural innovations and performance improvements. This survey presents a systematic review of deep learning-based VIS methods, introducing a novel categorization based on temporal modeling strategies: frame-by-frame, clip-based, in-memory feature propagation, and in-memory object query propagation. Comprehensive quantitative comparisons of existing work across three major VIS benchmark datasets are also provided. Additionally, emerging challenges in the field are explored, with several promising research directions identified, aiming to provide valuable insights for researchers and practitioners interested in VIS, while further advancing deep learning techniques for VIS. • Categorization of VIS approaches based on their temporal modeling strategies. • Comprehensive quantitative comparison of current VIS methods. • Analysis of the challenges and potential future research directions in VIS.
Cited by 5Open access
O, Nithina, Kumar P V, Prasanth · Zenodo (CERN European Organization for Nuclear Research) · 2015
Object recognition is the task of finding and identifying objects in an image or video sequence. Object categorization is a typical task of computer vision which involves determining whether or not an image contains some specific category of object. The idea is related with recognition, identification, and detection. The object recognition problem can be defined as a labeling problem based on models of known objects. This is closely tied to the segmentation problem. Without at least a partial recognition of objects, segmentation cannot be done, and without segmentation, object recognition is not possible. Object recognition is generally posed as the problem of matching a representation of the target object with the available image features, while rejecting the background features. This paper compares various image segmentation methods and recognition of categorized objects from noisy web image collection using auto-context model.
Cited by 0Open access
Antonio Torralba, Aude Oliva · Network Computation in Neural Systems · 2003
In this paper we study the statistical properties of natural images belonging to different categories and their relevance for scene and object categorization tasks. We discuss how second-order statistics are correlated with image categories, scene scale and objects. We propose how scene categorization could be computed in a feedforward manner in order to provide top-down and contextual information very early in the visual processing chain. Results show how visual categorization based directly on low-level features, without grouping or segmentation stages, can benefit object localization and identification. We show how simple image statistics can be used to predict the presence and absence of objects in the scene before exploring the image.
Cited by 794
Anubha Kale, Anurag Jain · 2014
Abstract — Image segmentation is an important and difficult task of image processing and the consequent tasks including object detection, feature extraction, object recognition and categorization depend on the quality of segmentation process. In this paper we suggest Genetic Algorithm to solve the problem of image segmentation. The problem was treated as optimization problems based GA. GAs is used to segment an image by using an optimization function without any threshold values. GAs based image segmentation can provide more accurate results than traditional segmentation methods. The genetic procedure provided a faster convergence to the optimal solution. This is because the sampling strategy allows exploring the solution space by a strategy that is not biased. The preliminary results indicate that GA-based methods perform better than the traditional methods in terms of quality. Also, by developing hybrid algorithms such as GAs and Artificial Neural Networks (ANNs) we can reduce the processing time and increase the visual quality of the final segmentation underscoring the advantages of hybrid algorithms.
Cited by 7
2008
Abstract — This paper presents a novel method for detecting and localizing objects of a visual category in cluttered real-world scenes. Our approach considers object categorization and figureground segmentation as two interleaved processes that closely collaborate towards a common goal. As shown in our work, the tight coupling between those two processes allows them to benefit from each other and improve the combined performance. The core part of our approach is a highly flexible learned representation for object shape that can combine the information observed on different training examples in a probabilistic extension of the Generalized Hough Transform. The resulting approach can detect categorical objects in novel images and automatically infer a probabilistic segmentation from the recognition result. This segmentation is then in turn used to again improve recognition by allowing the system to focus its efforts on object pixels and to discard misleading influences from the background. Moreover, the information from where in the image a hypothesis draws its support is employed in an MDL based hypothesis verification stage to resolve ambiguities between overlapping hypotheses and factor out the effects of partial occlusion. An extensive evaluation on several large data sets shows that the proposed system is applicable to a range of different object categories, including both rigid and articulated objects. In addition, its flexible representation allows it to achieve competitive object detection performance already from training sets that are between one and two orders of magnitude smaller than those used in comparable systems. Index Terms — object categorization, object detection, segmentation, clustering, Hough transform, hypothesis selection, MDL I.
Cited by 0
Mandy J. Maguire, Jonathan S. Brumberg, Michelle D. Ennis et al. · Spatial Cognition and Computation · 2011
Abstract Events, like objects, can be decomposed into parts. Path, the spatiotemporal trajectory of an object during an event, is the most commonly labeled event feature across the world's languages, provides important social information, and is increasingly central to theories of general event segmentation. However, little is understood about how adults visually segment paths. We apply theories developed for object segmentation to help understand path segmentation. Overall subjects segmented equivalent object shapes and event paths in similar ways following patterns predicted by Singh and Hoffman's (2001) Singh, M. and Hoffman, D. D. 2001. “Parts-based representations of visual shape and implications for visual cognition.”. In From fragments to objects—Segmentation and grouping in vision Edited by: Shipley, T. F. and Kellman, P. J. 401–459. New York, NY: Elsevier Science.. [Google Scholar] geometric analysis of object parts. There were two notable differences between object and event segmentation: (1) event parsing occurred at points of negative curvature minima and positive curvature maxima as opposed to simply negative curvature minima; and (2) event parsing was more frequent and variable than object parsing. Implications of these results for event perception and categorization are discussed.
Cited by 15
Kaiqiong Sun · IETE Technical Review · 2011
AbstractVascular angiogram is an important tool for diagnosing the presence of vascular disease, and its segmentation plays a key role for the objective and precise quantitative analysis of vascular tree shape and dimension. This review summarizes the current research efforts in the area of image segmentation of vessel angiogram, especially 2D projective angiogram, and presents some of the challenges anticipated in the years to come. Instead of a full description of the various methods based on categorization, the focus of this review is on providing the fundamental principles and relations of main segmentation methods for angiogram in terms of the image analysis aspect. The paper analyzes the different methods, such as local methods, tracking-based methods, and model-based methods, from a consistent aspect. The segmentation is summed up as a regular transform process and the different methods are aimed at a uniform segmentation object.
Cited by 13
Ferran Cabezas Castellví, Amaia Salvador · 2015
This thesis explores processing techniques to deal with noisy data in crowdsourced object segmentation tasks. It is used the data collected with Click’n’Cut, an online interactive segmentation tool, and it is performed several experiments towards improving the segmentation results. First, it is introduced different superpixel-based techniques to filter users ’ traces, and assess their impact in the segmentation result. Second, it is presented different criteria to detect and discard the traces from potential bad users, resulting in a remarkable increase in performance. Then, it is shown a novel superpixel-based segmentation algorithm which does not require any prior filtering and is based on weighting each user’s contribution according to his/her level of expertise. Finally, it is exposed different features and their corresponding rules for automatic categorizing the crowd users. The main application of this automatic categorization is to observe which pattern follow each user that produces bad traces and convert these traces into traces that give us better performance in the segmentation 2
Cited by 0
Dheerendra Pratap Singh, Manohar Yadav · International Journal of Remote Sensing · 2024
Point cloud has emerged as the most popular three-dimensional (3D) data format in recent years for several scientific and industrial applications. Point cloud semantic segmentation has piqued the researcher's interest, which is a crucial stage in 3D analysis and scene comprehension. Deep learning-based processing is more feasible to increase the availability of point cloud acquisition tools that is LiDAR systems at the user end. The point cloud learning achieves tremendous success in object detection, object categorization, and semantic segmentation. To summarize the recent works with chronological development, comprehensive review of projection-, voxel-, and direct point-based point cloud semantic segmentation methods is performed from various perspectives. The commonly used point cloud benchmark datasets with their characteristics are discussed, and they are used for the performance analysis and comparison of several state-of-the-art segmentation methods. The quantitative performance analysis of these deep learning models summarizes the trend of semantic segmentation of point clouds. In the context of point cloud semantic segmentation, the various methods have specific roles. Based on the review of methods working and their performance analysis, it is concluded that the projection-based methods prioritize efficiency, which is ideal in unavailability of high-performance computing system. Voxel-based methods capture overall context, serving well in 3D object classification. Point-based approaches excel in fine details and efficiency, suited for tasks like 3D semantic segmentation. Choosing the suitable method depends on the task, data, and resources. KPConv and DGCNN are popular choices, especially for precision and adaptability to point density. However, method performance varies, underlining the need for tailored selection. Hybrid approaches, combining method strengths, promise superior results.
Cited by 25
Kar Seng Loke · Monash University · 2017
Computer image understanding of pictorial information has many useful applications and is one of most researched areas in computer science. The process of automated image understanding often requires that the image is first automatically segmented. Image segmentation divides an image into coherent parts and has been predominantly performed by grouping regions of similarity or by partitioning the image based on edge detection. This form of segmentation is frequently unsatisfactory because it fails to segment at meaningful object boundaries. Therefore, our work is aimed at segmenting objects by extracting (or detecting) their boundaries. The segmentation of object boundaries and their identification can be considered as the first step towards semantic image scene understanding. In this work we introduce a new image representation that combines wedgelets and discrete orthogonal polynomials to model straight edges and textures. We use an Ant Colony Optimization (ACO) algorithm to search for object contours that satisfy local constraints using local cues. The ACO algorithm performs a biased exploration in alternation with optimization allowing us to escape local maxima. However, the results show that local cues do not provide sufficient information to the Ant algorithm to enable it to construct viable object contours. To overcome the local cue limitations, we examine the use of global cues, specifically the shape of the object being segmented, but this leads us to a potential paradox. If the purpose of segmentation is to obtain the identity of the object, then how can we obtain the shape of the object without first knowing its identity? We resolve this apparent paradox by using other methods to first identify the object and then use the result of the identification to guide the segmentation. Based on this concept we have developed a segmentation algorithm which first identifies the object using image patches and then segments the object using the active contour algorithm. The algorithm is rotation, translation and scale invariant. However, using image-based cues for recognition has limitations for general categorization, as many objects are usually categorized by their shape. We solve this by developing a shape-based cue for object recognition that is biologically inspired. Shape-based cues allow us to recognize objects based on their global shape. Our experimental results show that our method is comparable or better than known methods. We then use the edge-based cues to construct the objects contour detection using the ACO algorithm developed earlier. In this sense our work combines a top-down, bottom-up segmentation approach that has parallels in human visual systems.
Cited by 0Open access
Kar Seng Loke · Figshare · 2017
Computer image understanding of pictorial information has many useful applications and is one of most researched areas in computer science. The process of automated image understanding often requires that the image is first automatically segmented. Image segmentation divides an image into coherent parts and has been predominantly performed by grouping regions of similarity or by partitioning the image based on edge detection. This form of segmentation is frequently unsatisfactory because it fails to segment at meaningful object boundaries. Therefore, our work is aimed at segmenting objects by extracting (or detecting) their boundaries. The segmentation of object boundaries and their identification can be considered as the first step towards semantic image scene understanding. In this work we introduce a new image representation that combines wedgelets and discrete orthogonal polynomials to model straight edges and textures. We use an Ant Colony Optimization (ACO) algorithm to search for object contours that satisfy local constraints using local cues. The ACO algorithm performs a biased exploration in alternation with optimization allowing us to escape local maxima. However, the results show that local cues do not provide sufficient information to the Ant algorithm to enable it to construct viable object contours. To overcome the local cue limitations, we examine the use of global cues, specifically the shape of the object being segmented, but this leads us to a potential paradox. If the purpose of segmentation is to obtain the identity of the object, then how can we obtain the shape of the object without first knowing its identity? We resolve this apparent paradox by using other methods to first identify the object and then use the result of the identification to guide the segmentation. Based on this concept we have developed a segmentation algorithm which first identifies the object using image patches and then segments the object using the active contour algorithm. The algorithm is rotation, translation and scale invariant. However, using image-based cues for recognition has limitations for general categorization, as many objects are usually categorized by their shape. We solve this by developing a shape-based cue for object recognition that is biologically inspired. Shape-based cues allow us to recognize objects based on their global shape. Our experimental results show that our method is comparable or better than known methods. We then use the edge-based cues to construct the objects contour detection using the ACO algorithm developed earlier. In this sense our work combines a top-down, bottom-up segmentation approach that has parallels in human visual systems.
Cited by 0Open access
Kar Seng Loke · Monash University · 2017
Computer image understanding of pictorial information has many useful applications and is one of most researched areas in computer science. The process of automated image understanding often requires that the image is first automatically segmented. Image segmentation divides an image into coherent parts and has been predominantly performed by grouping regions of similarity or by partitioning the image based on edge detection. This form of segmentation is frequently unsatisfactory because it fails to segment at meaningful object boundaries. Therefore, our work is aimed at segmenting objects by extracting (or detecting) their boundaries. The segmentation of object boundaries and their identification can be considered as the first step towards semantic image scene understanding. In this work we introduce a new image representation that combines wedgelets and discrete orthogonal polynomials to model straight edges and textures. We use an Ant Colony Optimization (ACO) algorithm to search for object contours that satisfy local constraints using local cues. The ACO algorithm performs a biased exploration in alternation with optimization allowing us to escape local maxima. However, the results show that local cues do not provide sufficient information to the Ant algorithm to enable it to construct viable object contours. To overcome the local cue limitations, we examine the use of global cues, specifically the shape of the object being segmented, but this leads us to a potential paradox. If the purpose of segmentation is to obtain the identity of the object, then how can we obtain the shape of the object without first knowing its identity? We resolve this apparent paradox by using other methods to first identify the object and then use the result of the identification to guide the segmentation. Based on this concept we have developed a segmentation algorithm which first identifies the object using image patches and then segments the object using the active contour algorithm. The algorithm is rotation, translation and scale invariant. However, using image-based cues for recognition has limitations for general categorization, as many objects are usually categorized by their shape. We solve this by developing a shape-based cue for object recognition that is biologically inspired. Shape-based cues allow us to recognize objects based on their global shape. Our experimental results show that our method is comparable or better than known methods. We then use the edge-based cues to construct the objects contour detection using the ACO algorithm developed earlier. In this sense our work combines a top-down, bottom-up segmentation approach that has parallels in human visual systems.
Cited by 0Open access
Thomas Oskam · 2007
There exist many approaches to object recognition of image data. Most of the methods use a topdown approach to classify the content, as low-level information is often considered inapplicable or insufficient for this task. The method presented in this work, however, shows a way to exploit low-level image segmentation for the purpose of categorizing different object classes of still images. The key idea is to not only use one single segmentation, but a whole set of different parameterized segmentations of an image as basis for feature extraction. Segment boundaries are used as paths along which strings of feature vectors are drawn. These strings are pairwise aligned to create a scoring matrix, which can be transformed into a Mercer kernel for a standard SVM based classifier. Despite the inherit problems of low-level segmentations, this method performs very well on standard benchmark image databases and shows that segmentation indeed can be used for object recognition. The first goal of this semester thesis was to create a Matlab framework that implements the proposed method for object recognition. With this framework as basis, several approaches to improve scoring results of pairwise alignments, and thus improve the categorization rate, need
Cited by 0