A discriminatively trained, multiscale, deformable part model

By Pedro F. Felzenszwalb, David A. McAllester, Deva Ramanan. 2008

3,130 citations recorded by the source corpus.

Subfields: Artificial Intelligence, 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 Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Intriguing properties of neural networks (2013, background): “We make the connection with hard-negative mining explicitly, as it is close in spirit: hard-negative mining, in computer vision, consists of identifying training set examples (or portions thereof) which are given low probabilities by the model, but which should be high probability instead, cf. [5].”
Framing Image Description as a Ranking Task: Data, Models and Evaluation Metrics (2013, background): “…these systems (the only exception being Gupta et al., 2012) rely on detectors trained Hodosh, Young & Hockenmaier on images from the same data set (Felzenszwalb et al., 2008), it is unclear how well these approaches would generalize to other domains where no labeled data to train detectors is…”
SSD: Single Shot MultiBox Detector (2015, methodology): “Before the advent of convolutional neural networks, the state of the art for those two approaches – Deformable Part Model (DPM) [26] and Selective Search [1] – had comparable performance.”
Towards Accurate Multi-person Pose Estimation in the Wild (2017, methodology): “oneered by the Pictorial Structures model of Fischer and Elschlager [16]. One of the ﬁrst practical and well performing methods based on this idea is Deformable Part Model (DPM) by Felzenswalb et al. [15], which spurred a large body of work on probabilistic graphical models (PGM) for 2-D human pose ”
Prioritized Experience Replay (2015, methodology): “ly often, because their errors shrink less fast, and the chosen samples from the common classes will be those nearest to the decision boundaries, leading to an effect similar to hard negative mining (Felzenszwalb et al., 2008). To check whether these intuitions hold, we conducted a preliminary exper”
Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps (2013, methodology): “0% localisation error using a fully-supervised algorithm based on the part-based models [6] and Fisher vector feature encoding [11].”
Joint Training of a Convolutional Network and a Graphical Model for Human Pose Estimation (2014, methodology): “In a similar spirit, more general techniques using “Pictorial Structures” such as Felzenszwalb and colleagues work [12] made this approach tractable with so called ‘Deformable Part Models (DPM)’.”
Stacked Hourglass Networks for Human Pose Estimation (2016, background): “Early work tackles such diﬃculties using robust image features and sophisticated structured prediction [1,2,3,4,5,6,7,8,9]: the former is used to produce local interpretations, whereas the latter is used to infer a globally consistent pose.”
Uncovering and Mitigating Algorithmic Bias through Learned Latent Structure (2019, methodology): “Skin tone annotations are based on the Fitzpatrick skin type classiﬁcation system (Fitzpatrick 1988), with each image labeled as “Lighter” or “Darker”.”
Unsupervised Discovery of Mid-Level Discriminative Patches (2012, background): “, [14]), our discriminative patches can be discovered in a fully unsupervised manner – given only a large pile of unlabeled images.”
Efficient object localization using Convolutional Networks (2014, methodology): “Following the seminal work of Felzenszwalb et al. [10] on ‘Deformable Part Models’ (DPM) for human-body-pose estimation, many algorithms have been proposed to improve on the DPM archiecture [2, 9, 23, 7].”
Render for CNN: Viewpoint Estimation in Images Using CNNs Trained with Rendered 3D Model Views (2015, methodology): “3K chair models respectively to render tens of thousands training images, which are then used to train deformable part models (DPM [8]).”
DSSD : Deconvolutional Single Shot Detector (2017, background): “While sliding window approaches never completely disappeared, they had gone out of favor after the heydays of HOG [4] and DPM [7] due to the increasingly large number of box locations that had to be considered to keep up with state-of-the-art.”
Structural-RNN: Deep Learning on Spatio-Temporal Graphs (2015, methodology): “They have been applied to many applications [34, 13, 45] including st-graphs which are commonly modeled as spatio-temporal CRF [39, 33, 65, 11].”
Discovering objects and their relations from entangled scene representations (2017, background): “d in various branches of computer science, biology, and other ﬁelds. In computer vision, Zhu et al. (2009) and Zhao & Zhu (2011) modelled relations among image features using stochastic grammars, Felzenszwalb et al. (2008) modelled relations among object parts using the “deformable parts” model,”
R-CNN minus R (2015, neutral): “However, they diﬀer in part-based detector such as [7] in that they do not explicitly model object parts in learning; instead parts are implicitly capture in the CNN. Integrated training of SPP-CNN as a single CNN learning problem, not dissimilar to some of the ideas of Section 3.2, have very…”
Cut, Paste and Learn: Surprisingly Easy Synthesis for Instance Detection (2017, background): “Our key insight is that state-of-the art detection meth-ods like Faster-RCNN [39] and even older approaches like DPM [9] etc . care more about local region-based features for detection than the global scene layout.”
An empirical study of context in object detection (2009, methodology): “Therefore the rest of our analysis is conducted using the UoCTTI’07 detector on the VOC’08 validation set.”
Adversarial Diversity and Hard Positive Generation (2016, background): “∗This work supported in part by NSF#1320956 RI: Small: Open Vision A deeper problem is that the generated adversarial images are relatively portable across different neural networks, which means that they are consistently misclassified by models of similar network architectures trained on varying…”
Learning Human Pose Estimation Features with Convolutional Networks (2013, methodology): “Following the methodology of Felzenszwalb et al. [15], at test time we run our model on images with only one person (351 images of the 1016 test examples).”