3,415 Hits in 9.1 sec

ProposalCLIP: Unsupervised Open-Category Object Proposal Generation via Exploiting CLIP Cues [article]

Hengcan Shi, Munawar Hayat, Yicheng Wu, Jianfei Cai
2022 arXiv   pre-print
Finally, we present a proposal regression module that extracts pseudo labels based on CLIP cues and trains a lightweight network to further refine proposals.  ...  Secondly, a graph-based merging module is proposed to solve the limitations of CLIP cues and merge fragmented proposals.  ...  To extract pseudo labels, we leverage the first three parts in our method to generate proposals.  ... 
arXiv:2201.06696v1 fatcat:hk7urhquczbhxh62tlpjmckloe

3D Multi-Object Tracking Using Graph Neural Networks with Cross-Edge Modality Attention [article]

Martin Buchner, Abhinav Valada
2022 arXiv   pre-print
In this work, we propose Batch3DMOT which follows the tracking-by-detection paradigm and represents real-world scenes as directed, acyclic, and category-disjoint tracking graphs that are attributed using  ...  We present a multi-modal graph neural network that uses a cross-edge attention mechanism mitigating modality intermittence, which translates into sparsity in the graph domain.  ...  This demonstrates the efficacy of using pseudo-labels for training Kalman filters, based on data statistics.  ... 
arXiv:2203.10926v2 fatcat:4uxt4oupvzabzh254nn5et6t4i

RWSeg: Cross-graph Competing Random Walks for Weakly Supervised 3D Instance Segmentation [article]

Shichao Dong, Ruibo Li, Jiacheng Wei, Fayao Liu, Guosheng Lin
2022 arXiv   pre-print
RWSeg can generate qualitative instance-level pseudo labels.  ...  Furthermore, we propose a Cross-graph Competing Random Walks (CGCRW) algorithm which encourages competition among different instance graphs to resolve ambiguities in closely placed objects and improve  ...  In this work, we follow the annotation method used in SegGroup [36] and "One Thing One Click" [29] , which only requires one object to be annotated by one point on it.  ... 
arXiv:2208.05110v2 fatcat:btoybedkj5dsvcul54wmnzavsu

Open-World Instance Segmentation: Exploiting Pseudo Ground Truth From Learned Pairwise Affinity [article]

Weiyao Wang, Matt Feiszli, Heng Wang, Jitendra Malik, Du Tran
2022 arXiv   pre-print
Open-world instance segmentation is the task of grouping pixels into object instances without any pre-determined taxonomy.  ...  From PA we construct a large set of pseudo-ground-truth instance masks; combined with human-annotated instance masks we train GGNs and significantly outperform the SOTA on open-world instance segmentation  ...  We thank Ross Girshick for the discussion on baselines and grouping methods and Abhijit Ogale for the discussion about open-world setting.  ... 
arXiv:2204.06107v1 fatcat:4cshp6zubfgnhdsurhkz574o3a

Semi-supervised 3D Object Detection via Temporal Graph Neural Networks [article]

Jianren Wang, Haiming Gang, Siddarth Ancha, Yi-Ting Chen, David Held
2022 arXiv   pre-print
3D object detection plays an important role in autonomous driving and other robotics applications.  ...  Instead, we propose leveraging large amounts of unlabeled point cloud videos by semi-supervised learning of 3D object detectors via temporal graph neural networks.  ...  Our solution is to leverage uncertainty in the pseudo labels.  ... 
arXiv:2202.00182v1 fatcat:n7fptukzh5aabhvsrhtaj4ufle

Important Object Identification with Semi-Supervised Learning for Autonomous Driving [article]

Jiachen Li and Haiming Gang and Hengbo Ma and Masayoshi Tomizuka and Chiho Choi
2022 arXiv   pre-print
We propose a novel approach for important object identification in egocentric driving scenarios with relational reasoning on the objects in the scene.  ...  Our approach also outperforms rule-based baselines by a large margin.  ...  Object Importance Annotation: In order to obtain the binary importance labels of the object in each bounding box, a group of annotators (i.e., experienced drivers) were asked to watch the egocentric driving  ... 
arXiv:2203.02634v1 fatcat:w5ekhfs6ircxto6rt3nx76th6m

Uncertainty-aware Mean Teacher for Source-free Unsupervised Domain Adaptive 3D Object Detection [article]

Deepti Hegde, Vishwanath Sindagi, Velat Kilic, A. Brinton Cooper, Mark Foster, Vishal Patel
2021 arXiv   pre-print
Pseudo-label based self training approaches are a popular method for source-free unsupervised domain adaptation.  ...  Leveraging model uncertainty allows the mean teacher network to perform implicit filtering by down-weighing losses corresponding uncertain pseudo-labels.  ...  We evaluate the pseudo labels by comparing the pseudo class annotation with the ground truth annotation for each detected object.  ... 
arXiv:2109.14651v1 fatcat:5wo4dotrb5ejtnqcgqneagnkyi

Improve Event Extraction via Self-Training with Gradient Guidance [article]

Zhiyang Xu, Lifu Huang
2022 arXiv   pre-print
The new event predictions along with their correctness scores are then used as pseudo labeled examples to improve the base event extraction model while the magnitude and direction of its gradients are  ...  when the high-quality AMR graph annotations are not available.  ...  detected and labeled by the EE system.  ... 
arXiv:2205.12490v1 fatcat:qtg5hy732ffrflnej2ad2f34a4

Improving Generalizability of Graph Anomaly Detection Models via Data Augmentation [article]

Shuang Zhou, Xiao Huang, Ninghao Liu, Fu-Lai Chung, Long-Kai Huang
2022 arXiv   pre-print
In practice, people usually need to identify anomalies on new (sub)graphs to secure their business, but they may lack labels to train an effective detection model.  ...  In this paper, we base on the phenomenon and propose a general and novel research problem of generalized graph anomaly detection that aims to effectively identify anomalies on both the training-domain  ...  In short, for each labeled anomaly, we try to find its corresponding S h and generate h new as pseudo-labels, and we use S r to denote all the generated pseudo-labels. C.  ... 
arXiv:2209.10168v1 fatcat:66524vwysfdkzmhxbfwd7r4xse

Weak Novel Categories without Tears: A Survey on Weak-Shot Learning [article]

Li Niu
2022 arXiv   pre-print
In different tasks, weak annotations are presented in different forms (e.g., noisy labels for image classification, image labels for object detection, bounding boxes for segmentation), similar to the definitions  ...  Assuming the existence of base categories with adequate fully-annotated training samples, different paradigms requiring fewer training samples or weaker annotations for novel categories have attracted  ...  Weak-Shot Object Detection In weak-shot object detection, the base training samples have bounding boxes while the novel training samples only have image labels.  ... 
arXiv:2110.02651v3 fatcat:sscbggzjdvhtxbas6mjssgikq4

Unsupervised Domain Adaptation of Object Detectors: A Survey [article]

Poojan Oza, Vishwanath A. Sindagi, Vibashan VS, Vishal M. Patel
2021 arXiv   pre-print
Considering that detection is a fundamental task in computer vision, many recent works have focused on developing novel domain adaptive detection techniques.  ...  Here, we describe in detail the domain adaptation problem for detection and present an extensive survey of the various methods.  ...  [101] Pseudo-label self-training Arruda et al. [106] Zhang et al. [99] Graph reasoning Xu et al. [95] Zhao et al. [98] Sovinay et al. [123] Mean-teacher training Cai et al.  ... 
arXiv:2105.13502v2 fatcat:ozzbbvoflfdvjdewjnjmfajlpa

ClusterNet: unsupervised generic feature learning for fast interactive satellite image segmentation

Nicolas Girard, Andrii Zhygallo, Yuliya Tarabalka, Lorenzo Bruzzone, Francesca Bovolo, Jon Atli Benediktsson
2019 Image and Signal Processing for Remote Sensing XXV  
Those features are then used to leverage sparse human annotations to compute a dense segmentation of the image.  ...  The cross-entropy loss is used on the predicted labels and the pseudo-labels.  ...  with pseudo-labels L t as ground truth Output: Trained model M Our method for this step is based on DeepCluster 12 whose approach uses a CNN to predict a feature vector for an input image.  ... 
doi:10.1117/12.2532796 fatcat:5ice4izkkvekddzvicn3cp3isy

RefineLoc: Iterative Refinement for Weakly-Supervised Action Localization

Alejandro Pardo, Humam Alwassel, Fabian Caba Heilbron, Ali Thabet, Bernard Ghanem
2021 2021 IEEE Winter Conference on Applications of Computer Vision (WACV)  
To alleviate this problem, recent methods have tried to leverage weak labeling, where videos are untrimmed and only a video-level label is available.  ...  Video action detectors are usually trained using datasets with fully-supervised temporal annotations. Building such datasets is an expensive task.  ...  In the object detection domain, refining using pseudo ground truth considerably reduces the performance gap between fully and weakly-supervised object detectors [59, 71] .  ... 
doi:10.1109/wacv48630.2021.00336 fatcat:j3ubsot5rne5zbdwcdy2vxhojm

Maximize the Exploration of Congeneric Semantics for Weakly Supervised Semantic Segmentation [article]

Ke Zhang, Sihong Chen, Qi Ju, Yong Jiang, Yucong Li, Xin He
2021 arXiv   pre-print
To explore the congeneric semantic regions from the same class to the maximum, we construct the patch-level graph neural network (P-GNN) based on the self-detected patches from different images that contain  ...  The graph network that is established with patches as the nodes can maximize the mutual learning of similar objects.  ...  Another cropping method is like the object detection methods. Each patch is the detected region. Here we adopt the weakly supervised object detection method, which only uses class labels like ours.  ... 
arXiv:2110.03982v1 fatcat:juz5t7m6vrferdixhhfodsehq4

TokenCut: Segmenting Objects in Images and Videos with Self-supervised Transformer and Normalized Cut [article]

Yangtao Wang
2022 arXiv   pre-print
In this paper, we describe a graph-based algorithm that uses the features obtained by a self-supervised transformer to detect and segment salient objects in images and videos.  ...  Detection and segmentation of salient objects is then formulated as a graph-cut problem and solved using the classical Normalized Cut algorithm.  ...  We have investigated whether such learned features can be used with a graph-based approach to detect and segment salient objects in images and videos (Fig. 1b ), formulating the segmentation problem using  ... 
arXiv:2209.00383v2 fatcat:rxlnpbguaze35nioydv6js3khi
« Previous Showing results 1 — 15 out of 3,415 results