Weakly Supervised Detection with Posterior Regularization

Bilen, Hakan; Pedersoli, Marco; Tuytelaars, Tinne

doi:10.5244/c.28.52

Cited by 79 publications

(142 citation statements)

References 0 publications

Supporting

Mentioning

141

Contrasting

Order By: Relevance

“…In practice, however, the quality of a solution depends on the similarity measure used. For instance, [52, 10, 12] obtain a solution set that is most consistent in terms of shape and color, [14,15] ex-165 ploit motion and appearance consistency within the input data and [11] exploits symmetry constraints of objects in a multiclass framework. The solution is mostly obtained by multiple instance learning [10] or by minimizing an energy on a fully connected graph [14,19].…”

Section: Related Workmentioning

confidence: 99%

“…This is a suboptimal choice because all other instances of that object in the image are ignored therefore failing to tap its true potential. This limitation is dealt in [11] by introducing a latent SVM 175 formulation that exploits presence of multiple object instances in an image. On similar lines, the present method is a generalization of [19] where the assumption of selecting strictly one instance per video is relaxed in the framework of exploiting human context for building models of small 180 and medium sized objects.…”

Section: Related Workmentioning

confidence: 99%

“…Most 155 approaches impose certain assumptions for successful application e.g. [12,11,10,52,14] assume a single predominant object in the input data and [14,15] assume rigid or articulated objects with motion distinctive from its background. These assumptions guide the latent variables such 160 that the solution extracts object instances despite object deformations and background.…”

Section: Related Workmentioning

confidence: 99%

“…Moreover, this also ignores vast amount of freely available weakly structured data. As a result, recent works in object detection have turned towards utilizing weakly labelled data [5, 6,7,8,9,10,11,12], particularly 20 videos [13,14,15]. Critically, these methods assume that motion or appearance of objects are sufficient descriptors for segmenting them with relative ease, which is indeed the case for large active objects such as flying airplanes and walking tigers.…”

Section: Introductionmentioning

confidence: 98%

See 3 more Smart Citations

Weak supervision for detecting object classes from activities

Srikantha

Gall

2017

Computer Vision and Image Understanding

View full text Add to dashboard Cite

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 98%

See 2 more Smart Citations

Weak supervision for detecting object classes from activities

Srikantha

Gall

2017

Computer Vision and Image Understanding

View full text Add to dashboard Cite

“…Bilen et al [25] incorporate domain-specific prior knowledge (such as mutual exclusion and symmetry) in a posterior regularization setting with a softmax margin learning framework for object detection. Even though this approach can handle a limited number of multiple objects in each image, it requires careful formulation of the prior knowledge and the optimization is challenging and can be sensitive to local optima.…”

Section: Related Workmentioning

confidence: 99%

Efficient Labelling of Pedestrian Supervisions

Htike

2016

ELCVIA

View full text Add to dashboard Cite

Object detection is a fundamental goal to achieve intelligent visual perception by computers due to the fact that objects are the basic building blocks to achieve higher level image understanding. Among the numerous categories of objects in the real-world, pedestrians are among the most important due to several potential benefits brought about by successful pedestrian detection. Often, pedestrian detectors are trained in state-of-the-art systems using supervised machine learning algorithms which necessitates costly and often tedious manual annotation of pedestrians in the form of precise bounding boxes. In this paper, a novel weakly supervised learning algorithm is proposed to train a pedestrian detector that requires, instead of bounding boxes, only annotations of estimated centres of pedestrians. The algorithm makes use of a pedestrian prior learnt in an unsupervised way from the input video and this prior is fused with the given weak supervision information in a systematic manner. By evaluating on publicly available datasets, we demonstrate that our weakly supervised algorithm reduces the cost of manual annotation of pedestrians by more than four times while achieving similar performance to a pedestrian detector trained with standard bounding box annotations.

show abstract

Zero-Annotation Object Detection with Web Knowledge Transfer

Tao

Yang

Cai

2018

Computer Vision – ECCV 2018

View full text Add to dashboard Cite

Object detection is one of the major problems in computer vision, and has been extensively studied. Most of the existing detection works rely on labor-intensive supervision, such as ground truth bounding boxes of objects or at least image-level annotations. On the contrary, we propose an object detection method that does not require any form of human annotation on target tasks, by exploiting freely available web images. In order to facilitate effective knowledge transfer from web images, we introduce a multi-instance multi-label domain adaption learning framework with two key innovations. First of all, we propose an instance-level adversarial domain adaptation network with attention on foreground objects to transfer the object appearances from web domain to target domain. Second, to preserve the class-specific semantic structure of transferred object features, we propose a simultaneous transfer mechanism to transfer the supervision across domains through pseudo strong label generation. With our end-to-end framework that simultaneously learns a weakly supervised detector and transfers knowledge across domains, we achieved significant improvements over baseline methods on the benchmark datasets.

show abstract

Weakly Supervised Detection with Posterior Regularization

Cited by 79 publications

References 0 publications

Weak supervision for detecting object classes from activities

Weak supervision for detecting object classes from activities

Efficient Labelling of Pedestrian Supervisions

Zero-Annotation Object Detection with Web Knowledge Transfer

Contact Info

Product

Resources

About