You searched for +publisher:"Colorado State University" +contributor:("Beveridge, J. Ross").
Showing records 1 – 14 of
14 total matches.
No search limiters apply to these results.
Colorado State University
1.
Mulay, Gururaj.
Adapting RGB pose estimation to new domains.
Degree: MS(M.S.), Computer Science, 2019, Colorado State University
URL: http://hdl.handle.net/10217/195409 
► Many multi-modal human computer interaction (HCI) systems interact with users in real-time by estimating the user's pose. Generally, they estimate human poses using depth sensors…
(more)
▼ Many multi-modal human computer interaction (HCI) systems interact with users in real-time by estimating the user's pose. Generally, they estimate human poses using depth sensors such as the Microsoft Kinect.For multi-modal HCI interfaces to gain traction in the real world, however, it would be better for pose estimation to be based on data from RGB cameras, which are more common and less expensive than depth sensors. This has motivated research into pose estimation from RGB images. Convolutional Neural Networks (CNNs) represent the
state-of-the-art in this literature, for example [1–5], and [6]. These systems estimate 2D human poses from RGB images. A problem with current CNN-based pose estimators is that they require large amounts of labeled data for training. If the goal is to train an RGB pose estimator for a new domain, the cost of collecting and more importantly labeling data can be prohibitive. A common solution is to train on publicly available pose data sets, but then the trained system is not tailored to the domain. We propose using RGB+D sensors to collect domain-specific data in the lab, and then training the RGB pose estimator using skeletons automatically extracted from the RGB+D data. This paper presents a case study of adapting the RMPE pose estimation network [4] to the domain of the DARPA Communicating with Computers (CWC) program [7], as represented by the EGGNOG data set [8]. We chose RMPE because it predicts both joint locations and Part Affinity Fields (PAFs) in real-time. Our adaptation of RMPE trained on automatically-labeled data outperforms the original RMPE on the EGGNOG data set.
Advisors/Committee Members: Draper, Bruce (advisor), Beveridge, J. Ross (advisor), Maciejewsky, Anthony (committee member).
Subjects/Keywords: CWC; human pose estimation; RMPE; HCI; convolutional neural networks; Microsoft Kinect
Record Details
Similar Records
Cite
Share »
Record Details
Similar Records
Cite
« Share
❌
APA ·
Chicago ·
MLA ·
Vancouver ·
CSE |
Export
to Zotero / EndNote / Reference
Manager
APA (6th Edition):
Mulay, G. (2019). Adapting RGB pose estimation to new domains. (Masters Thesis). Colorado State University. Retrieved from http://hdl.handle.net/10217/195409
Chicago Manual of Style (16th Edition):
Mulay, Gururaj. “Adapting RGB pose estimation to new domains.” 2019. Masters Thesis, Colorado State University. Accessed February 27, 2021.
http://hdl.handle.net/10217/195409.
MLA Handbook (7th Edition):
Mulay, Gururaj. “Adapting RGB pose estimation to new domains.” 2019. Web. 27 Feb 2021.
Vancouver:
Mulay G. Adapting RGB pose estimation to new domains. [Internet] [Masters thesis]. Colorado State University; 2019. [cited 2021 Feb 27].
Available from: http://hdl.handle.net/10217/195409.
Council of Science Editors:
Mulay G. Adapting RGB pose estimation to new domains. [Masters Thesis]. Colorado State University; 2019. Available from: http://hdl.handle.net/10217/195409
Colorado State University
2.
D'Souza, Wimroy.
Evaluating the role of context in 3D theater stage reconstruction.
Degree: MS(M.S.), Electrical and Computer Engineering, 2014, Colorado State University
URL: http://hdl.handle.net/10217/88518
► Recovering the 3D structure from 2D images is a problem dating back to the 1960s. It is only recently, with the advancement of computing technology,…
(more)
▼ Recovering the 3D structure from 2D images is a problem dating back to the 1960s. It is only recently, with the advancement of computing technology, that there has been substantial progress in solving this problem. In this thesis, we focus on one method for recovering scene structure given a single image. This method uses supervised learning techniques and a multiple-segmentation framework for adding contextual information to the inference. We evaluate the effect of this added contextual information by excluding this additional information to measure system performance. We then go on to evaluate the effect of the other system components that remain which include classifiers and image features. For example, in the case of classifiers, we substitute the original with others to see the level of accuracy that these provide. In the case of the features, we conduct experiments that give us the most important features that contribute to classification accuracy. All of this put together lets us evaluate the effect of adding contextual information to the learning process and if it can be improved by improving the other non-contextual components of the system.
Advisors/Committee Members: Beveridge, J. Ross (advisor), Draper, Bruce (committee member), Luo, J. Rockey (committee member).
Subjects/Keywords: 3D theater stage reconstruction; scene understanding; computer vision
Record Details
Similar Records
Cite
Share »
Record Details
Similar Records
Cite
« Share
❌
APA ·
Chicago ·
MLA ·
Vancouver ·
CSE |
Export
to Zotero / EndNote / Reference
Manager
APA (6th Edition):
D'Souza, W. (2014). Evaluating the role of context in 3D theater stage reconstruction. (Masters Thesis). Colorado State University. Retrieved from http://hdl.handle.net/10217/88518
Chicago Manual of Style (16th Edition):
D'Souza, Wimroy. “Evaluating the role of context in 3D theater stage reconstruction.” 2014. Masters Thesis, Colorado State University. Accessed February 27, 2021.
http://hdl.handle.net/10217/88518.
MLA Handbook (7th Edition):
D'Souza, Wimroy. “Evaluating the role of context in 3D theater stage reconstruction.” 2014. Web. 27 Feb 2021.
Vancouver:
D'Souza W. Evaluating the role of context in 3D theater stage reconstruction. [Internet] [Masters thesis]. Colorado State University; 2014. [cited 2021 Feb 27].
Available from: http://hdl.handle.net/10217/88518.
Council of Science Editors:
D'Souza W. Evaluating the role of context in 3D theater stage reconstruction. [Masters Thesis]. Colorado State University; 2014. Available from: http://hdl.handle.net/10217/88518
Colorado State University
3.
McNeely-White, David G.
Same data, same features: modern ImageNet-trained convolutional neural networks learn the same thing.
Degree: MS(M.S.), Computer Science, 2020, Colorado State University
URL: http://hdl.handle.net/10217/208467
► Deep convolutional neural networks (CNNs) are the dominant technology in computer vision today. Much of the recent computer vision literature can be thought of as…
(more)
▼ Deep convolutional neural networks (CNNs) are the dominant technology in computer vision today. Much of the recent computer vision literature can be thought of as a competition to find the best architecture for vision within the deep convolutional framework. Despite all the effort invested in developing sophisticated convolutional architectures, however, it's not clear how different from each other the best CNNs really are. This thesis measures the similarity between ten well-known CNNs, in terms of the properties they extract from images. I find that the properties extracted by each of the ten networks are very similar to each other, in the sense that any of their features can be well approximated by an affine transformation of the features of any of the other nine. In particular, there is evidence that each network extracts mostly the same information as each other network, though some do it more robustly. The similarity between each of these CNNs is surprising. Convolutional neural networks learn complex non-linear features of images, and the architectural differences between systems suggest that these non-linear functions should take different forms. Nonetheless, these ten CNNs which were trained on the same data set seem to have learned to extract similar properties from images. In essence, each CNN's training algorithm hill-climbs in a very different parameter space, yet converges on a similar solution. This suggests that for CNNs, the selection of the training set and strategy may be more important than the selection of the convolutional architecture.
Advisors/Committee Members: Beveridge, J. Ross (advisor), Anderson, Charles W. (committee member), Seger, Carol A. (committee member).
Subjects/Keywords: convolutional neural networks; feature space; machine learning; feature mapping; computer vision; ImageNet
Record Details
Similar Records
Cite
Share »
Record Details
Similar Records
Cite
« Share
❌
APA ·
Chicago ·
MLA ·
Vancouver ·
CSE |
Export
to Zotero / EndNote / Reference
Manager
APA (6th Edition):
McNeely-White, D. G. (2020). Same data, same features: modern ImageNet-trained convolutional neural networks learn the same thing. (Masters Thesis). Colorado State University. Retrieved from http://hdl.handle.net/10217/208467
Chicago Manual of Style (16th Edition):
McNeely-White, David G. “Same data, same features: modern ImageNet-trained convolutional neural networks learn the same thing.” 2020. Masters Thesis, Colorado State University. Accessed February 27, 2021.
http://hdl.handle.net/10217/208467.
MLA Handbook (7th Edition):
McNeely-White, David G. “Same data, same features: modern ImageNet-trained convolutional neural networks learn the same thing.” 2020. Web. 27 Feb 2021.
Vancouver:
McNeely-White DG. Same data, same features: modern ImageNet-trained convolutional neural networks learn the same thing. [Internet] [Masters thesis]. Colorado State University; 2020. [cited 2021 Feb 27].
Available from: http://hdl.handle.net/10217/208467.
Council of Science Editors:
McNeely-White DG. Same data, same features: modern ImageNet-trained convolutional neural networks learn the same thing. [Masters Thesis]. Colorado State University; 2020. Available from: http://hdl.handle.net/10217/208467
Colorado State University
4.
Dragan, Matthew R.
Demonstrating that dataset domains are largely linearly separable in the feature space of common CNNs.
Degree: MS(M.S.), Computer Science, 2020, Colorado State University
URL: http://hdl.handle.net/10217/219571
► Deep convolutional neural networks (DCNNs) have achieved state of the art performance on a variety of tasks. These high-performing networks require large and diverse training…
(more)
▼ Deep convolutional neural networks (DCNNs) have achieved
state of the art performance on a variety of tasks. These high-performing networks require large and diverse training datasets to facilitate generalization when extracting high-level features from low-level data. However, even with the availability of these diverse datasets, DCNNs are not prepared to handle all the data that could be thrown at them. One major challenges DCNNs face is the notion of forced choice. For example, a network trained for image classification is configured to choose from a predefined set of labels with the expectation that any new input image will contain an instance of one of the known objects. Given this expectation it is generally assumed that the network is trained for a particular domain, where domain is defined by the set of known object classes as well as more implicit assumptions that go along with any data collection. For example, some implicit characteristics of the ImageNet dataset domain are that most images are taken outdoors and the object of interest is roughly in the center of the frame. Thus the domain of the network is defined by the training data that is chosen. Which leads to the following key questions: Does a network know the domain it was trained for? and Can a network easily distinguish between in-domain and out-of-domain images? In this thesis it will be shown that for several widely used public datasets and commonly used neural networks, the answer to both questions is yes. The presence of a simple method of differentiating between in-domain and out-of-domain cases has significant implications for work on domain adaptation, transfer learning, and model generalization.
Advisors/Committee Members: Beveridge, J. Ross (advisor), Ortega, Francisco (committee member), Peterson, Chris (committee member).
Record Details
Similar Records
Cite
Share »
Record Details
Similar Records
Cite
« Share
❌
APA ·
Chicago ·
MLA ·
Vancouver ·
CSE |
Export
to Zotero / EndNote / Reference
Manager
APA (6th Edition):
Dragan, M. R. (2020). Demonstrating that dataset domains are largely linearly separable in the feature space of common CNNs. (Masters Thesis). Colorado State University. Retrieved from http://hdl.handle.net/10217/219571
Chicago Manual of Style (16th Edition):
Dragan, Matthew R. “Demonstrating that dataset domains are largely linearly separable in the feature space of common CNNs.” 2020. Masters Thesis, Colorado State University. Accessed February 27, 2021.
http://hdl.handle.net/10217/219571.
MLA Handbook (7th Edition):
Dragan, Matthew R. “Demonstrating that dataset domains are largely linearly separable in the feature space of common CNNs.” 2020. Web. 27 Feb 2021.
Vancouver:
Dragan MR. Demonstrating that dataset domains are largely linearly separable in the feature space of common CNNs. [Internet] [Masters thesis]. Colorado State University; 2020. [cited 2021 Feb 27].
Available from: http://hdl.handle.net/10217/219571.
Council of Science Editors:
Dragan MR. Demonstrating that dataset domains are largely linearly separable in the feature space of common CNNs. [Masters Thesis]. Colorado State University; 2020. Available from: http://hdl.handle.net/10217/219571
Colorado State University
5.
Bolme, David Scott.
Theory and applications of optimized correlation output filters.
Degree: PhD, Computer Science, 2011, Colorado State University
URL: http://hdl.handle.net/10217/47326
► Correlation filters are a standard way to solve many problems in signal processing, image processing, and computer vision. This research introduces two new filter training…
(more)
▼ Correlation filters are a standard way to solve many problems in signal processing, image processing, and computer vision. This research introduces two new filter training techniques, called Average of Synthetic Exact Filters (ASEF) and Minimum Output Sum of Squared Error (MOSSE), which have produced filters that perform well on many object detection problems. Typically, correlation filters are created by cropping templates out of training images; however, these templates fail to adequately discriminate between targets and background in difficult detection scenarios. More advanced methods such as Synthetic Discriminant Functions (SDF), Minimum Average Correlation Energy (MACE), Unconstrained Minimum Average Correlation Energy (UMACE), and Optimal Tradeoff Filters (OTF) improve performance by controlling the response of the correlation peak, but they only loosely control the effect of the filters on the rest of the image. This research introduces a new approach to correlation filter training, which considers the entire image to image mapping known as cross-correlation. ASEF and MOSSE find filters that optimally map the input training images to user specified outputs. The goal is to produce strong correlation peaks for targets while suppressing the responses to background. Results in eye localization, person detection, and visual tracking indicate that these new filters outperform other advanced correlation filter training methods and even produce better results than much more complicated non-filter algorithms.
Advisors/Committee Members: Beveridge, J. Ross, 1957- (advisor), Draper, Bruce A. (Bruce Austin), 1962- (committee member), Strout, Michelle Mills (committee member), Kirby, Michael J. (committee member).
Subjects/Keywords: computer vision; object detection; correlation filters
Record Details
Similar Records
Cite
Share »
Record Details
Similar Records
Cite
« Share
❌
APA ·
Chicago ·
MLA ·
Vancouver ·
CSE |
Export
to Zotero / EndNote / Reference
Manager
APA (6th Edition):
Bolme, D. S. (2011). Theory and applications of optimized correlation output filters. (Doctoral Dissertation). Colorado State University. Retrieved from http://hdl.handle.net/10217/47326
Chicago Manual of Style (16th Edition):
Bolme, David Scott. “Theory and applications of optimized correlation output filters.” 2011. Doctoral Dissertation, Colorado State University. Accessed February 27, 2021.
http://hdl.handle.net/10217/47326.
MLA Handbook (7th Edition):
Bolme, David Scott. “Theory and applications of optimized correlation output filters.” 2011. Web. 27 Feb 2021.
Vancouver:
Bolme DS. Theory and applications of optimized correlation output filters. [Internet] [Doctoral dissertation]. Colorado State University; 2011. [cited 2021 Feb 27].
Available from: http://hdl.handle.net/10217/47326.
Council of Science Editors:
Bolme DS. Theory and applications of optimized correlation output filters. [Doctoral Dissertation]. Colorado State University; 2011. Available from: http://hdl.handle.net/10217/47326
Colorado State University
6.
Teli, Mohammad Nayeem.
Face detection using correlation filters.
Degree: PhD, Computer Science, 2013, Colorado State University
URL: http://hdl.handle.net/10217/80983
► Cameras are ubiquitous and available all around us. As a result, images and videos are posted online in huge numbers. These images often need to…
(more)
▼ Cameras are ubiquitous and available all around us. As a result, images and videos are posted online in huge numbers. These images often need to be stored and analyzed. This requires the use of various computer vision applications that includes detection of human faces in these images and videos. The emphasis on face detection is evident from the applications found in everyday point and shoot cameras for a better focus, on social networking sites for tagging friends and family and for security situations which subsequently require face recognition or verification. This thesis focuses on detecting human faces in still images and video frames using correlation filters. These correlation filters are trained using a recent technique called Minimum Output Sum of Squared Error (MOSSE) developed by Bolme et al. Since correlation filters identify only a peak location, it only helps in localizing a single target point. In this thesis, I develop techniques to use this localization for detection of human faces of different scales and poses in uncontrolled background, location and lighting conditions. The goal of this research is to extend correlation filters for face detection and identify the scenarios where its potential is the most. The specific contributions of this work are the development of a novel face detector using correlation filters and the identification of the strengths and weaknesses of this approach. This approach is applied to an easy dataset and a hard dataset to emphasize the efficacy of correlations filters for face detection. This technique shows 95.6% accuracy in finding the exact location of the faces in images with controlled background and lighting. Although, the results on a hard dataset were not better than the OpenCV Viola and Jones face detector, it showed much better results, 81.5% detection rate compared to 69.43% detection rate by the Viola and Jones face detector, when tested on a customized dataset that was controlled for location change between training and test datasets. This result signifies the strength of a correlation based face detector in a specific scenario with uniform setting, such as a building entrance or an airport security gate.
Advisors/Committee Members: Beveridge, J. Ross (advisor), Draper, Bruce A. (committee member), Howe, Adele (committee member), Givens, Geof H. (committee member).
Subjects/Keywords: MOSSE; face detection; point and shoot; correlation filters; face
Record Details
Similar Records
Cite
Share »
Record Details
Similar Records
Cite
« Share
❌
APA ·
Chicago ·
MLA ·
Vancouver ·
CSE |
Export
to Zotero / EndNote / Reference
Manager
APA (6th Edition):
Teli, M. N. (2013). Face detection using correlation filters. (Doctoral Dissertation). Colorado State University. Retrieved from http://hdl.handle.net/10217/80983
Chicago Manual of Style (16th Edition):
Teli, Mohammad Nayeem. “Face detection using correlation filters.” 2013. Doctoral Dissertation, Colorado State University. Accessed February 27, 2021.
http://hdl.handle.net/10217/80983.
MLA Handbook (7th Edition):
Teli, Mohammad Nayeem. “Face detection using correlation filters.” 2013. Web. 27 Feb 2021.
Vancouver:
Teli MN. Face detection using correlation filters. [Internet] [Doctoral dissertation]. Colorado State University; 2013. [cited 2021 Feb 27].
Available from: http://hdl.handle.net/10217/80983.
Council of Science Editors:
Teli MN. Face detection using correlation filters. [Doctoral Dissertation]. Colorado State University; 2013. Available from: http://hdl.handle.net/10217/80983
Colorado State University
7.
Kulkarni, Hrushikesh N.
Performance evaluation of feature sets for carried object detection in still images.
Degree: MS(M.S.), Electrical and Computer Engineering, 2014, Colorado State University
URL: http://hdl.handle.net/10217/83983
► Human activity recognition has gathered a lot of interest. The ability to accurately detect carried objects on human beings will directly help activity recognition. This…
(more)
▼ Human activity recognition has gathered a lot of interest. The ability to accurately detect carried objects on human beings will directly help activity recognition. This thesis performs evaluation of four different features for carried object detection. To detect carried objects, image chips in a video are extracted by tracking moving objects using an off the shelf tracker. Pixels with similar colors are grouped together by using a superpixel segmentation algorithm. Features are calculated with respect to every superpixel, encoding information regarding their location in the track chip, shape of the superpixel, pose of the person in the track chip, and appearance of the superpixel. ROC curves are used for analyzing the detection of a superpixel as a carried object using these features individually or in a combination. These ROC curves show that the detection using Shape features as they are calculated have very less information. The location features, though simple to calculate, have a significant usable information. Detection using pose of a person in the track chip and appearance of the superpixel depend largely on the data used for their calculation. Pose detections are more likely to be correct if there are no occlusions, while appearance work better if we have high resolution of input images.
Advisors/Committee Members: Beveridge, J. Ross (advisor), Draper, Bruce A. (advisor), Pasricha, Sudeep (committee member), Alciatore, David G. (committee member).
Subjects/Keywords: carried object detection; computer vision; image processing; machine learning; performance evaluation of features; video surveillance
Record Details
Similar Records
Cite
Share »
Record Details
Similar Records
Cite
« Share
❌
APA ·
Chicago ·
MLA ·
Vancouver ·
CSE |
Export
to Zotero / EndNote / Reference
Manager
APA (6th Edition):
Kulkarni, H. N. (2014). Performance evaluation of feature sets for carried object detection in still images. (Masters Thesis). Colorado State University. Retrieved from http://hdl.handle.net/10217/83983
Chicago Manual of Style (16th Edition):
Kulkarni, Hrushikesh N. “Performance evaluation of feature sets for carried object detection in still images.” 2014. Masters Thesis, Colorado State University. Accessed February 27, 2021.
http://hdl.handle.net/10217/83983.
MLA Handbook (7th Edition):
Kulkarni, Hrushikesh N. “Performance evaluation of feature sets for carried object detection in still images.” 2014. Web. 27 Feb 2021.
Vancouver:
Kulkarni HN. Performance evaluation of feature sets for carried object detection in still images. [Internet] [Masters thesis]. Colorado State University; 2014. [cited 2021 Feb 27].
Available from: http://hdl.handle.net/10217/83983.
Council of Science Editors:
Kulkarni HN. Performance evaluation of feature sets for carried object detection in still images. [Masters Thesis]. Colorado State University; 2014. Available from: http://hdl.handle.net/10217/83983
Colorado State University
8.
Kadappan, Karthik.
Element rearrangement for action classification on product manifolds.
Degree: MS(M.S.), Electrical and Computer Engineering, 2013, Colorado State University
URL: http://hdl.handle.net/10217/80250
► Conventional tensor-based classification algorithms unfold tensors into matrices using the standard mode-k unfoldings and perform classification using established machine learning algorithms. These methods assume that…
(more)
▼ Conventional tensor-based classification algorithms unfold tensors into matrices using the standard mode-k unfoldings and perform classification using established machine learning algorithms. These methods assume that the standard mode-k unfolded matrices are the best 2-dimensional representations of N-dimensional structures. In this thesis, we ask the question: "Is there a better way to unfold a tensor?" To address this question, we design a method to create unfoldings of a tensor by rearranging elements in the original tensor and then applying the standard mode-k unfoldings. The rearrangement of elements in a tensor is formulated as a combinatorial optimization problem and tabu search is adapted in this work to solve it. We study this element rearrangement problem in the context of tensor-based action classification on product manifolds. We assess the proposed methods using a publicly available video data set, namely Cambridge-Gesture data set. We design several neighborhood structures and search strategies for tabu search and analyze their performance. Results reveal that the proposed element rearrangement algorithm developed in this thesis can be employed as a preprocessing step to increase classification accuracy in the context of action classification on product manifolds method.
Advisors/Committee Members: Beveridge, J. Ross (advisor), Maciejewski, Anthony A. (committee member), Peterson, Chris (committee member), Rajopadhye, Sanjay (committee member).
Subjects/Keywords: action classification; computer vision; element rearrangement; manifolds; Tabu search; tensor
Record Details
Similar Records
Cite
Share »
Record Details
Similar Records
Cite
« Share
❌
APA ·
Chicago ·
MLA ·
Vancouver ·
CSE |
Export
to Zotero / EndNote / Reference
Manager
APA (6th Edition):
Kadappan, K. (2013). Element rearrangement for action classification on product manifolds. (Masters Thesis). Colorado State University. Retrieved from http://hdl.handle.net/10217/80250
Chicago Manual of Style (16th Edition):
Kadappan, Karthik. “Element rearrangement for action classification on product manifolds.” 2013. Masters Thesis, Colorado State University. Accessed February 27, 2021.
http://hdl.handle.net/10217/80250.
MLA Handbook (7th Edition):
Kadappan, Karthik. “Element rearrangement for action classification on product manifolds.” 2013. Web. 27 Feb 2021.
Vancouver:
Kadappan K. Element rearrangement for action classification on product manifolds. [Internet] [Masters thesis]. Colorado State University; 2013. [cited 2021 Feb 27].
Available from: http://hdl.handle.net/10217/80250.
Council of Science Editors:
Kadappan K. Element rearrangement for action classification on product manifolds. [Masters Thesis]. Colorado State University; 2013. Available from: http://hdl.handle.net/10217/80250
9.
Alsaaran, Hessah.
Unsupervised video segmentation using temporal coherence of motion.
Degree: PhD, Computer Science, 2015, Colorado State University
URL: http://hdl.handle.net/10217/170396
► Spatio-temporal video segmentation groups pixels with the goal of representing moving objects in scenes. It is a difficult task for many reasons: parts of an…
(more)
▼ Spatio-temporal video segmentation groups pixels with the goal of representing moving objects in scenes. It is a difficult task for many reasons: parts of an object may look very different from each other, while parts of different objects may look similar and/or overlap. Of particular importance to this dissertation, parts of non-rigid objects such as animals may move in different directions at the same time. While appearance models are good for segmenting visually distinct objects and traditional motion models are good for segmenting rigid objects, there is a need for a new technique to segment objects that move non-rigidly. This dissertation presents a new unsupervised motion-based video segmentation approach. It segments non-rigid objects based on motion temporal coherence (i.e. the correlations of when points move), instead of motion magnitude and direction as in previous approaches. The hypothesis is that although non-rigid objects can move their parts in different directions, their parts tend to move at the same time. In the experiments, the proposed approach achieves better results than related
state-of-the-art approaches on a video of zebras in the wild, and on 41 videos from the VSB100 dataset.
Advisors/Committee Members: Draper, Bruce A. (advisor), Beveridge, J. Ross (advisor), Whitley, Darrell (committee member), Peterson, Christopher (committee member).
Subjects/Keywords: temporal coherence of motion; video segmentation
Record Details
Similar Records
Cite
Share »
Record Details
Similar Records
Cite
« Share
❌
APA ·
Chicago ·
MLA ·
Vancouver ·
CSE |
Export
to Zotero / EndNote / Reference
Manager
APA (6th Edition):
Alsaaran, H. (2015). Unsupervised video segmentation using temporal coherence of motion. (Doctoral Dissertation). Colorado State University. Retrieved from http://hdl.handle.net/10217/170396
Chicago Manual of Style (16th Edition):
Alsaaran, Hessah. “Unsupervised video segmentation using temporal coherence of motion.” 2015. Doctoral Dissertation, Colorado State University. Accessed February 27, 2021.
http://hdl.handle.net/10217/170396.
MLA Handbook (7th Edition):
Alsaaran, Hessah. “Unsupervised video segmentation using temporal coherence of motion.” 2015. Web. 27 Feb 2021.
Vancouver:
Alsaaran H. Unsupervised video segmentation using temporal coherence of motion. [Internet] [Doctoral dissertation]. Colorado State University; 2015. [cited 2021 Feb 27].
Available from: http://hdl.handle.net/10217/170396.
Council of Science Editors:
Alsaaran H. Unsupervised video segmentation using temporal coherence of motion. [Doctoral Dissertation]. Colorado State University; 2015. Available from: http://hdl.handle.net/10217/170396
10.
Patil, Dhruva.
Looking under the hood: visualizing what LSTMs learn.
Degree: MS(M.S.), Computer Science, 2019, Colorado State University
URL: http://hdl.handle.net/10217/197280
► Recurrent Neural Networks (RNNs) such as Long Short Term Memory (LSTM) and Gated Recurrent Units (GRUs) have been successful in many applications involving sequential data.…
(more)
▼ Recurrent Neural Networks (RNNs) such as Long Short Term Memory (LSTM) and Gated Recurrent Units (GRUs) have been successful in many applications involving sequential data. The success of these models lies in the complex feature representations they learn from the training data. One criteria to trust the model is its validation accuracy. However, this can lead to surprises when the network learns properties of the input data, different from what the designer intended and/or the user assumes. As a result, we lack confidence in even high-performing networks when they are deployed in applications with novel input data, or where the cost of failure is very high. Thus understanding and visualizing what recurrent networks have learned becomes essential. Visualizations of RNN models are better established in the field of natural language processing than in computer vision. This work presents visualizations of what recurrent networks, particularly LSTMs, learn in the domain of action recognition, where the inputs are sequences of 3D human poses, or skeletons. The goal of the thesis is to understand the properties learned by a network with regard to an input action sequence, and how it will generalize to novel inputs. This thesis presents two methods for visualizing concepts learned by RNNs in the domain of action recognition, providing an independent insight into the working of the recognition model. The first visualization method shows the sensitivity of joints over time in a video sequence. The second visualization method generates synthetic videos that maximize the responses of a class label or hidden unit within a set of known anatomical constraints. These techniques are combined in a visualization tool called SkeletonVis to help developers and users gain insights into models embedded in RNNs for action recognition. We present case studies on NTU-RGBD, a popular data set for action recognition, to reveal properties learnt by a trained LSTM network.
Advisors/Committee Members: Draper, Bruce (advisor), Beveridge, J. Ross (committee member), Maciejewski, Anthony (committee member).
Subjects/Keywords: activation maximization; recurrent neural networks; action recognition; visualization; LSTM
Record Details
Similar Records
Cite
Share »
Record Details
Similar Records
Cite
« Share
❌
APA ·
Chicago ·
MLA ·
Vancouver ·
CSE |
Export
to Zotero / EndNote / Reference
Manager
APA (6th Edition):
Patil, D. (2019). Looking under the hood: visualizing what LSTMs learn. (Masters Thesis). Colorado State University. Retrieved from http://hdl.handle.net/10217/197280
Chicago Manual of Style (16th Edition):
Patil, Dhruva. “Looking under the hood: visualizing what LSTMs learn.” 2019. Masters Thesis, Colorado State University. Accessed February 27, 2021.
http://hdl.handle.net/10217/197280.
MLA Handbook (7th Edition):
Patil, Dhruva. “Looking under the hood: visualizing what LSTMs learn.” 2019. Web. 27 Feb 2021.
Vancouver:
Patil D. Looking under the hood: visualizing what LSTMs learn. [Internet] [Masters thesis]. Colorado State University; 2019. [cited 2021 Feb 27].
Available from: http://hdl.handle.net/10217/197280.
Council of Science Editors:
Patil D. Looking under the hood: visualizing what LSTMs learn. [Masters Thesis]. Colorado State University; 2019. Available from: http://hdl.handle.net/10217/197280
Colorado State University
11.
Stevens, John.
Analysis of grating cell features for texture discrimination, An.
Degree: MS(M.S.), Computer Science, 2010, Colorado State University
URL: http://hdl.handle.net/10217/41470
► The design of artificial vision systems has been influenced by knowledge of the early stages of processing in the human vision system. The discovery of…
(more)
▼ The design of artificial vision systems has been influenced by knowledge of the early stages of processing in the human vision system. The discovery of directionally sensitive cells in the human visual cortex lead to the theory of edge detection in computer vision, and the discovery that simple cell receptive fields can be modeled as Gabor filters has led to the development and use of Gabor jets. In this thesis, we evaluate a low-level image feature inspired by "grating" cells found in the human visual cortex. These cells, and the features based on them, detect spatial gratings–repeated patterns of light and dark bars–in their receptive fields. We evaluate the utility of grating cell model features to distinguish different textures using Fisher’s linear discriminant. It will be shown that the grating cell features contain significantly more distinguishing information than another standard Gabor-filter-based image feature.
Advisors/Committee Members: Draper, Bruce A. (Bruce Austin), 1962- (advisor), Troup, Lucy (advisor), Beveridge, J. Ross, 1957- (committee member).
Subjects/Keywords: Image analysis; Imaging systems; Computer vision
Record Details
Similar Records
Cite
Share »
Record Details
Similar Records
Cite
« Share
❌
APA ·
Chicago ·
MLA ·
Vancouver ·
CSE |
Export
to Zotero / EndNote / Reference
Manager
APA (6th Edition):
Stevens, J. (2010). Analysis of grating cell features for texture discrimination, An. (Masters Thesis). Colorado State University. Retrieved from http://hdl.handle.net/10217/41470
Chicago Manual of Style (16th Edition):
Stevens, John. “Analysis of grating cell features for texture discrimination, An.” 2010. Masters Thesis, Colorado State University. Accessed February 27, 2021.
http://hdl.handle.net/10217/41470.
MLA Handbook (7th Edition):
Stevens, John. “Analysis of grating cell features for texture discrimination, An.” 2010. Web. 27 Feb 2021.
Vancouver:
Stevens J. Analysis of grating cell features for texture discrimination, An. [Internet] [Masters thesis]. Colorado State University; 2010. [cited 2021 Feb 27].
Available from: http://hdl.handle.net/10217/41470.
Council of Science Editors:
Stevens J. Analysis of grating cell features for texture discrimination, An. [Masters Thesis]. Colorado State University; 2010. Available from: http://hdl.handle.net/10217/41470
Colorado State University
12.
Yamaguchi, Takanobu.
Cloud-top entrainment analyzed with a Lagrangian parcel tracking model in large-eddy simulations.
Degree: PhD, Atmospheric Science, 2010, Colorado State University
URL: http://hdl.handle.net/10217/44769
► Despite decades of research, cloud-top entrainment has not been described with firm evidence. This leads to insufficient understanding of the physics of marine stratocumulus clouds.…
(more)
▼ Despite decades of research, cloud-top entrainment has not been described with firm evidence. This leads to insufficient understanding of the physics of marine stratocumulus clouds. A Lagrangian Parcel Tracking Model (LPTM) was implemented in a large-eddy simulation model for detailed and direct analysis of the entrained air parcel following the parcel trajectory. The scalar advection scheme of the host model was replaced by a monotonic multidimensional odd-order conservative advection scheme. Tests with an idealized scalar field and stratocumulus turbulence suggested that the fifth-order scheme is optimal. Evaluation of the LPTM was performed with stratocumulus simulations. Parcel statistics agreed with Eulerian statistics, and the parcel paths agreed with the theoretical parcel paths. The Lagrangian budget equation for a scalar, however, generally does not hold for a simulated turbulence field, since the fractal nature of turbulence may cause numerical errors. Two large-eddy simulations were performed with grid spacing of O (5 m). The power spectra of these runs showed relatively good agreement with the energy cascade slope. A comparison with low-resolution simulations suggested that horizontal refinement is necessary for better representation of entrainment and microphysical processes. The LPTM with the high-resolution stratocumulus simulation showed that the location of entrainment is in cloud holes, which are drier downdraft regions. Parcels in the inversion layer, subsiding from the free atmosphere, are entrained in to the mixed layer. They are cooled and moistened by radiation, evaporation, and mixing. A mixing fraction analysis shows that the coolings during entrainment due to radiation and evaporation are comparable. The largest contribution to buoyancy reduction is the cooling due to mixing, for our simulation. The analysis also shows that buoyancy reversal occurs for the entrained parcels. Radiative cooling and cloud-top entrainment instability (CTEI) interact such that the radiative cooling forces larger saturation mixing fractions while CTEI forces smaller values. Additional simulations suggest that radiative cooling produces a negative feedback on the entrainment rate, which is strong enough to control turbulence and hide CTEI. Under such conditions, cloud breakup due to CTEI is unlikely.
Advisors/Committee Members: Randall, David A., (David Allan), 1948- (advisor), Beveridge, J. Ross, 1957- (committee member), Denning, A. Scott (committee member), Krueger, Steven K. (committee member), Schubert, Wayne H. (committee member).
Subjects/Keywords: stratocumulus cloud; large eddy simulation; Lagrangian parcel tracking; entrainment; cloud-top entrainment instability; buoyancy reversal; Atmospheric turbulence; Cloud physics; Boundary layer (Meteorology); Turbulence
Record Details
Similar Records
Cite
Share »
Record Details
Similar Records
Cite
« Share
❌
APA ·
Chicago ·
MLA ·
Vancouver ·
CSE |
Export
to Zotero / EndNote / Reference
Manager
APA (6th Edition):
Yamaguchi, T. (2010). Cloud-top entrainment analyzed with a Lagrangian parcel tracking model in large-eddy simulations. (Doctoral Dissertation). Colorado State University. Retrieved from http://hdl.handle.net/10217/44769
Chicago Manual of Style (16th Edition):
Yamaguchi, Takanobu. “Cloud-top entrainment analyzed with a Lagrangian parcel tracking model in large-eddy simulations.” 2010. Doctoral Dissertation, Colorado State University. Accessed February 27, 2021.
http://hdl.handle.net/10217/44769.
MLA Handbook (7th Edition):
Yamaguchi, Takanobu. “Cloud-top entrainment analyzed with a Lagrangian parcel tracking model in large-eddy simulations.” 2010. Web. 27 Feb 2021.
Vancouver:
Yamaguchi T. Cloud-top entrainment analyzed with a Lagrangian parcel tracking model in large-eddy simulations. [Internet] [Doctoral dissertation]. Colorado State University; 2010. [cited 2021 Feb 27].
Available from: http://hdl.handle.net/10217/44769.
Council of Science Editors:
Yamaguchi T. Cloud-top entrainment analyzed with a Lagrangian parcel tracking model in large-eddy simulations. [Doctoral Dissertation]. Colorado State University; 2010. Available from: http://hdl.handle.net/10217/44769
Colorado State University
13.
Lui, Yui Man.
Geometric methods on special manifolds for visual recognition.
Degree: PhD, Computer Science, 2010, Colorado State University
URL: http://hdl.handle.net/10217/39042
► Many computer vision methods assume that the underlying geometry of images is Euclidean. This assumption is generally not valid. Therefore, this dissertation introduces new nonlinear…
(more)
▼ Many computer vision methods assume that the underlying geometry of images is Euclidean. This assumption is generally not valid. Therefore, this dissertation introduces new nonlinear geometric frameworks based upon special manifolds, namely Graβmann and Stiefel manifolds, for visual recognition. The motivation for this thesis is driven by the intrinsic geometry of visual data in which the visual data can be either a still image or video. Visual data are represented as points in appropriately chosen parameter spaces. The idiosyncratic aspects of the data in these spaces are then exploited for pattern classification. Three major research results are presented in this dissertation: face recognition for illumination spaces on Stiefel manifolds, face recognition on Graβmann registration manifolds, and action classification on product manifolds. Previous work has shown that illumination cones are idiosyncratic for face recognition in illumination spaces. However, it has not been addressed how a single image relates to an illumination cone. In this dissertation, a Bayesian model is employed to relight a single image to a set of illuminated variants. The subspace formed by these illuminated variants is characterized on a Stiefel manifold. A new distance measure called Canonical Stiefel Quotient (CSQ) is introduced. CSQ performs two projections on a tangent space of a Stiefel manifold and uses the quotient for classification. The proposed method demonstrates that illumination cones can be synthesized by relighting a single image to a set of images, and the synthesized illumination cones are discriminative for face recognition. Experiments on the CMU-PIE and YaleB data sets reveal that CSQ not only achieves high recognition accuracies for generic faces but also is robust to the choice of training sets. Subspaces can be realized as points on Graβmann manifolds. Motivated by image perturbation and the geometry of Graβmann manifolds, we present a method called Graβmann Registration Manifolds (GRM) for face recognition. First, a tangent space is formed by a set of affine perturbed images where the tangent space admits a vector space structure. Second, the tangent spaces are embedded on a Graβmann manifold and chordal distance is used to compare subspaces. Experiments on the FERET database suggest that the proposed method yields excellent results using both holistic and local features. Specifically, on the FERET Dup2 data set, which is generally considered the most difficult data set on FERET, the proposed method achieves the highest rank one identification rate among all non-trained methods currently in the literature. Human actions compose a series of movements and can be described by a sequence of video frames. Since videos are multidimensional data, data tensors are the natural choice for data representation. In this dissertation, a data tensor is expressed as a point on a product manifold and classification is performed on this product space. First, we factorize a data tensor using a modified High Order Singular Value…
Advisors/Committee Members: Beveridge, J. Ross (advisor), Kirby, Michael, 1961- (committee member), Draper, Bruce A. (Bruce Austin), 1962- (committee member), Whitley, L. Darrell (committee member).
Subjects/Keywords: action classification; visual recognition; special manifolds; geometric methods; face recognition; Human face recognition (Computer science); Grassmann manifolds; Stiefel manifolds
Record Details
Similar Records
Cite
Share »
Record Details
Similar Records
Cite
« Share
❌
APA ·
Chicago ·
MLA ·
Vancouver ·
CSE |
Export
to Zotero / EndNote / Reference
Manager
APA (6th Edition):
Lui, Y. M. (2010). Geometric methods on special manifolds for visual recognition. (Doctoral Dissertation). Colorado State University. Retrieved from http://hdl.handle.net/10217/39042
Chicago Manual of Style (16th Edition):
Lui, Yui Man. “Geometric methods on special manifolds for visual recognition.” 2010. Doctoral Dissertation, Colorado State University. Accessed February 27, 2021.
http://hdl.handle.net/10217/39042.
MLA Handbook (7th Edition):
Lui, Yui Man. “Geometric methods on special manifolds for visual recognition.” 2010. Web. 27 Feb 2021.
Vancouver:
Lui YM. Geometric methods on special manifolds for visual recognition. [Internet] [Doctoral dissertation]. Colorado State University; 2010. [cited 2021 Feb 27].
Available from: http://hdl.handle.net/10217/39042.
Council of Science Editors:
Lui YM. Geometric methods on special manifolds for visual recognition. [Doctoral Dissertation]. Colorado State University; 2010. Available from: http://hdl.handle.net/10217/39042
Colorado State University
14.
Crawford-Hines, Stewart.
Machine learned boundary definitions for an expert's tracing assistant in image processing.
Degree: PhD, Computer Science, 2003, Colorado State University
URL: http://hdl.handle.net/10217/28552
► Most image processing work addressing boundary definition tasks embeds the assumption that an edge in an image corresponds to the boundary of interest in the…
(more)
▼ Most image processing work addressing boundary definition tasks embeds the assumption that an edge in an image corresponds to the boundary of interest in the world. In straightforward imagery this is true, however it is not always the case. There are images in which edges are indistinct or obscure, and these images can only be segmented by a human expert. The work in this dissertation addresses the range of imagery between the two extremes of those straightforward images and those requiring human guidance to appropriately segment. By freeing systems of a priori edge definitions and building in a mechanism to learn the boundary definitions needed, systems can do better and be more broadly applicable. This dissertation presents the construction of such a boundary-learning system and demonstrates the validity of this premise on real data. A framework was created for the task in which expert-provided boundary exemplars are used to create training data, which in turn are used by a neural network to learn the task and replicate the expert's boundary tracing behavior. This is the framework for the Expert's Tracing Assistant (ETA) system. For a representative set of nine structures in the Visible Human imagery, ETA was compared and contrasted to two
state-of-the-art, user guided methods – Intelligent Scissors (IS) and Active Contour Models (ACM). Each method was used to define a boundary, and the distances between these boundaries and an expert's ground truth were compared. Across independent trials, there will be a natural variation in an expert's boundary tracing, and this degree of variation served as a benchmark against which these three methods were compared. For simple structural boundaries, all the methods were equivalent. However, in more difficult cases, ETA was shown to significantly better replicate the expert's boundary than either IS or ACM. In these cases, where the expert's judgement was most called into play to bound the structure, ACM and IS could not adapt to the boundary character used by the expert while ETA could.
Advisors/Committee Members: Anderson, Charles W. (advisor), Draper, Bruce A. (Bruce Austin), 1962- (committee member), Beveridge, J. Ross (committee member), Alciatore, David G. (committee member).
Subjects/Keywords: visible human imagery; boundary definitions; expert's tracing assistant; ETA; intelligent scissors; IS; active contour models; ACM; boundary-learning system; Image processing; Pattern recognition systems
Record Details
Similar Records
Cite
Share »
Record Details
Similar Records
Cite
« Share
❌
APA ·
Chicago ·
MLA ·
Vancouver ·
CSE |
Export
to Zotero / EndNote / Reference
Manager
APA (6th Edition):
Crawford-Hines, S. (2003). Machine learned boundary definitions for an expert's tracing assistant in image processing. (Doctoral Dissertation). Colorado State University. Retrieved from http://hdl.handle.net/10217/28552
Chicago Manual of Style (16th Edition):
Crawford-Hines, Stewart. “Machine learned boundary definitions for an expert's tracing assistant in image processing.” 2003. Doctoral Dissertation, Colorado State University. Accessed February 27, 2021.
http://hdl.handle.net/10217/28552.
MLA Handbook (7th Edition):
Crawford-Hines, Stewart. “Machine learned boundary definitions for an expert's tracing assistant in image processing.” 2003. Web. 27 Feb 2021.
Vancouver:
Crawford-Hines S. Machine learned boundary definitions for an expert's tracing assistant in image processing. [Internet] [Doctoral dissertation]. Colorado State University; 2003. [cited 2021 Feb 27].
Available from: http://hdl.handle.net/10217/28552.
Council of Science Editors:
Crawford-Hines S. Machine learned boundary definitions for an expert's tracing assistant in image processing. [Doctoral Dissertation]. Colorado State University; 2003. Available from: http://hdl.handle.net/10217/28552
. 
Open Access Theses and Dissertations
