Weakly-supervised Discovery of Visual Pattern Configurations

By Hyun Oh Song, Yong Jae Lee, Stefanie Sabrina Jegelka, Trevor Darrell. 2014

166 citations recorded by the source corpus.

The prominence of weakly labeled data gives rise to a growing demand for object detection methods that can cope with minimal supervision. We propose an approach that automatically identifies discriminative configurations of visual patterns that are characteristic of a given object class. We formulate the problem as a constrained submodular optimization problem and demonstrate the benefits of the discovered configurations in remedying mislocalizations and finding informative positive and negative training examples. Together, these lead to state-of-the-art weakly-supervised detection results on the challenging PASCAL VOC dataset.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Fully Convolutional Multi-Class Multiple Instance Learning (2014, background): “MIL can reduce the need for bounding box annotations (Cinbis et al., 2014; Song et al., 2014), but it is rarely attempted for segmentation.”
Few-Shot Object Detection via Feature Reweighting (2018, background): “The weakly-supervised setting [3, 8, 40] considers the problem of training object detectors with only image-level labels, but without bounding box annotations, which are more expensive to obtain.”
Weakly Supervised Object Localization with Multi-Fold Multiple Instance Learning (2015, background): “’14 [47] 36.”
Weakly Supervised Object Localization and Detection: A Survey (2021, unknown): “Song et al. [137], [138] determine discriminative feature conﬁgurations of an object class via graph modeling, and train object detectors within the multiple-instance learning paradigm.”
Revisiting Knowledge Transfer for Training Object Class Detectors (2017, methodology): “We build on Multiple Instance Learning (MIL), a popular framework for weakly supervised object localization [27, 7, 2, 6, 9, 36, 38, 33], and extend it to incorporate knowledge from the source classes.”
Y Qualification Main task Y Return N N Y User qualified ? N Feedback Pass ? Pass ? Annotator training Annotating images (2017, methodology): “State-of-the-art WSOL approaches based on AlexNet architecture [34] perform as follows.”
Omni-DETR: Omni-Supervised Object Detection with Transformers (2022, unknown): “image-level tags [4, 17, 23, 42, 44] or instance-level points [8, 19, 32], and usually formulate WSOD as a multiple instance learning (MIL) problem.”
Dataset fingerprints: Exploring image collections through data mining (2015, methodology): “Our method is also somewhat related to recent weakly supervised methods such as [21, 22].”
Hide-and-Seek: Forcing a Network to be Meticulous for Weakly-Supervised Object and Action Localization (2017, methodology): “[43] combine multiple co-occurring discriminative regions to cover a larger extent of the object.”
Meta-Learning to Detect Rare Objects (2019, background): “There are other settings of detection with limited supervision, such as weakly-supervised detection [59, 5, 11], semi-supervised detection [69, 39, 12], and zero-shot detection [2, 41, 73], which are different from ours.”
Few-Shot Object Detection and Viewpoint Estimation for Objects in the Wild (2020, background): “Object detection with limited labeled samples is also addressed by approaches targeting weak supervision [49, 5, 6, 47] and zero-shot learning [3, 40, 66], but these settings are different from ours.”
Min-Entropy Latent Model for Weakly Supervised Object Detection (2018, background): “Bilen and Song [13], [14] leveraged clustering to initialize latent variables, i.e., object regions, part conﬁgurations and sub-categories, and learn object detectors based on the initialization.”
C-MIL: Continuation Multiple Instance Learning for Weakly Supervised Object Detection (2019, background): “WSOD approaches only require image category annotations indicating the presence or absence of a category of objects in images, significantly reducing human involvement by omitting labor-intensive bounding-box annotations [6,12,27–29,40].”
Deep multiple instance learning for image classification and auto-annotation (2015, methodology): “Recently, researchers started to study weakly supervised learning using features learned with deep representations [47, 35].”
Discover and Learn New Objects from Documentaries (2017, methodology): “The timeconsuming annotation process in object detection can be sidestepped by weakly supervised learning [2, 3, 7, 9, 13, 30, 33, 36, 40, 41, 42, 43, 44].”
W2F: A Weakly-Supervised to Fully-Supervised Framework for Object Detection (2018, methodology): “Most existing methods formulate weakly-supervised detection as an MIL problem [1, 31, 23, 16, 29, 18].”
Dissimilarity Coefficient Based Weakly Supervised Object Detection (2018, methodology): “Other end-to-end trainable frameworks for WSOD employ domain adaptation [22, 31], expectationmaximization algorithm [18, 39] and saliency based methods [21].”