Search Constraints

Start Over
Filtering by: Creator Oros, Daniel R. Remove constraint Creator: Oros, Daniel R. Advisor Li, Fuxin Remove constraint Advisor: Li, Fuxin

Search Results

  • Visual object discovery and understanding

    Creator:
    Yuan, Jialin
    Abstract:
    Learning to recognize objects is a fundamental and essential step in human perception and understanding of the world. Accordingly, research of object discovery across diverse modalities plays a pivotal role in the context of computer vision. This field not only contributes significantly to enhancing our understanding of visual information but...
    Resource Type:
    Dissertation
    Full Text:
    almost as good as mask R-CNN without depending on a two-stage framework. Note that the training of the

  • Toward Disentangling the Activations of the Deep Networks via Low-dimensional Embedding and Non-negative Factorization

    Creator:
    Khorram, Saeed
    Abstract:
    In this thesis, we introduce a novel Explanation Neural Network (XNN) to explain the predictions made by a deep network. The XNN works by embedding a high-dimensional activation vector of a deep network layer non-linearly into a low-dimensional explanation space while retaining faithfulness i.e., the original deep learning predictions can...
    Resource Type:
    Masters Thesis
    Full Text:
    rank r = 16. It can be observed that NMF basis vectors are more local and intuitive to humans while

  • Individual Fairness Optimization in Machine Learning with Minimax Loss and an Abstain Option

    Creator:
    Behnoudfar, Ali
    Abstract:
    In this thesis, a new learning algorithm is introduced that is targeted towards individual fairness. In order to be individually fair, mispredictions need to be avoided as each such prediction means the learning algorithm was unfair towards some individual. Therefore, achieving individual fairness implies having a perfect classifier, which is...
    Resource Type:
    Masters Thesis
    Full Text:
    } denotes the binary ground truth label. Assuming discriminant function f() and convex loss function φ : R 7

  • Web-based Deep Segmentation of Building Structure

    Creator:
    Chen, Zehuan
    Abstract:
    Deep learning and neural network has been widely used in research, deep learning has empowered many tasks such as point clouds segmentation and shape recognition. One of the main advantages of deep interaction point cloud segmentation is that it allows the feature extraction can be learned through neural network based...
    Resource Type:
    Masters Thesis
    Full Text:
    index) in the input cloud all neighbouring points within a sphere around Pi with the radius r are

  • Video Analysis : Techniques for Semi-Supervised Video Object Instance Segmentation and Tracking-by-Detection in the Wild

    Creator:
    Firl, Alrik K.
    Abstract:
    This thesis consists of two major components. The first part is concerned with video object instance segmentation (VOS), which is the task of assigning per-pixel labels perframe of a video sequence to indicate foreground object instance membership, given the first frame ground truth mask. VOS has myriad applications, from video...
    Resource Type:
    Masters Thesis
    Full Text:
    Faster R-CNN architecture diagram; we use Faster R-CNN with a ResNet- 101 feature extractor as one of

  • Leveraging Structures of the Data in Deep Learning

    Creator:
    Li, Xingyi
    Abstract:
    The performance of deep learning frameworks could be significantly improved through considering the particular underlying structures for each dataset. In this thesis, I summarize our three work about boosting the performance of deep learning models through leveraging structures of the data. In the first work, we theoretically justify that, for...
    Resource Type:
    Dissertation
    Full Text:
    trained with Faster R-CNN; (c) Sobolev-regularized cubic polynomial (Best Viewed in Color

  • Deep Perception Without a Camera: Enabling 3D Reconstruction and Object Recognition using Lidar and Sonar Sensing

    Creator:
    DeBortoli, Robert A.
    Abstract:
    Deep learning has recently revolutionized robot perception in many canonical robotic applications, such as autonomous driving. However, a similar transformation has yet to occur in more harsh environments including underwater and underground. This is due in part to the difficulty in deploying robots in these environments, which lack large real...
    Resource Type:
    Dissertation
    Full Text:
    ), respective synthetic sonar images (d)-(f), and reconstructed meshes (g)-(r). Ground truth point clouds are

  • Deep Convolutional Network on Point Clouds for 3D Scene Understanding

    Creator:
    Wu, Wenxuan
    Abstract:
    As one of the most popular data types, the point cloud is widely used in various appli- cations, including computer vision, computer graphics and robotics. The capability to directly measure 3D point clouds is invaluable in those applications as depth information could remove a lot of the segmentation ambiguities in...
    Resource Type:
    Dissertation
    Full Text:
    Details of the PointConvFormer Operation. h(pi − p) : R3 7→ Rk×cmid and ψ(Xpi − Xp) : Rcin 7→R are

  • Flex-shape convolution and an interactive image segmentation system

    Creator:
    Zhou, Zheng
    Abstract:
    Semantic image segmentation is a relatively difficult task in computer vision. With the advent of deep learning, semantic image segmentation is increasingly of interest for researchers because of the excellent predictions from Convolutional Neural Network (CNN). However, CNNs have proven to struggle with obtaining global context of image due to...
    Resource Type:
    Masters Thesis
    Full Text:
    for now on instance segmentation is Mask R-CNN[9] with 32.0% AP. Mask R-CNN is extended from Faster R

  • Towards Annotation-Efficient Video Representation Learning

    Creator:
    Patravali, Jay Sanjay
    Abstract:
    Labeling videos is costly, time-consuming and tedious. These costs can escalate in applications such as medical diagnosis or autonomous driving where we need domain expertise for annotation. Few-shot action recognition aims to solve this problem by annotation-efficient learning mechanisms. This thesis presents MetaUVFS as the first Unsupervised Meta-learning algorithm for...
    Resource Type:
    Masters Thesis
    Full Text:
    for query matching, and R-3DFSV [79] uses a large pre- trained 3D CNN along with weak labels to