12 Hits in 0.8 sec

Discriminatively Activated Sparselets

Ross B. Girshick, Hyun Oh Song, Trevor Darrell
2013 International Conference on Machine Learning  
In this paper we describe a new training framework that learns which sparselets to activate in order to optimize a discriminative objective, leading to larger speedup factors with no decrease in task performance  ...  Our experimental results demonstrate that discriminative activation substantially outperforms the previous reconstructive approach which, together with our structured output prediction formulation, make  ...  In Sec. 3, we describe how discriminative sparselet activation training fits into the framework and discuss several regularization methods for sparse activation learning.  ... 
dblp:conf/icml/GirshickSD13 fatcat:wfuhzxdfzvbgbhiqqk5gbvfwwm

Learning coarse-to-fine sparselets for efficient object detection and scene classification

Gong Cheng, Junwei Han, Lei Guo, Tianming Liu
2015 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)  
Then, we simultaneously train fine sparselets and activation vectors using a supervised single-hidden-layer neural network, in which sparselets training and discriminative activation vectors learning are  ...  constraint on the activation vectors.  ...  The experimental results in [2] and [3] showed that discriminatively activated sparselets outperform the previous sparse coding-based sparselets significantly.  ... 
doi:10.1109/cvpr.2015.7298721 dblp:conf/cvpr/ChengH0L15 fatcat:ia2fprq4nbbsjlliqge26wzbgu

Effective and Efficient Midlevel Visual Elements-Oriented Land-Use Classification Using VHR Remote Sensing Images

Gong Cheng, Junwei Han, Lei Guo, Zhenbao Liu, Shuhui Bu, Jinchang Ren
2015 IEEE Transactions on Geoscience and Remote Sensing  
To address this problem, we next propose a novel framework to train coarse-to-fine shared intermediate representations, which are termed "sparselets," from a large number of pretrained part detectors.  ...  with their corresponding discriminative activation vectors.  ...  Fig. 5 . 5 Framework of our presented coarse-to-fine sparselets training: (a) coarse sparselets training based on unsupervised single-hidden-layer auto-encoder; (b) fine sparselets and discriminative activation  ... 
doi:10.1109/tgrs.2015.2393857 fatcat:ydeelsiqurdntobxcyjhpvrrkq

Sparselet Models for Efficient Multiclass Object Detection [chapter]

Hyun Oh Song, Stefan Zickler, Tim Althoff, Ross Girshick, Mario Fritz, Christopher Geyer, Pedro Felzenszwalb, Trevor Darrell
2012 Lecture Notes in Computer Science  
Others optimize a discriminative objective [9, 10] . [11] builds a taxonomy of object classes based on shared features.  ...  --Ψ * D K --      = AM, (2) where M is a matrix of all sparselet responses, A is the matrix of sparse activation vectors.  ... 
doi:10.1007/978-3-642-33709-3_57 fatcat:ufzoyzmbk5el3buf55hruxh4ri

Fast, Accurate Detection of 100,000 Object Classes on a Single Machine

Thomas Dean, Mark A. Ruzon, Mark Segal, Jonathon Shlens, Sudheendra Vijayanarasimhan, Jay Yagnik
2013 2013 IEEE Conference on Computer Vision and Pattern Recognition  
We suspect it will be significant, and that our approach will complement the sparselet work.  ...  s method, spreading nowsparse filter activations into a more dense map using the part deformation scores. Maxima in this map are our object detections.  ... 
doi:10.1109/cvpr.2013.237 dblp:conf/cvpr/DeanRSSVY13 fatcat:meyjergglne37ph4uu2pdfrpky

The Fastest Deformable Part Model for Object Detection

Junjie Yan, Zhen Lei, Longyin Wen, Stan Z. Li
2014 2014 IEEE Conference on Computer Vision and Pattern Recognition  
A proximal gradient algorithm is adopted to progressively learn the low rank filter in a discriminative manner.  ...  Sparselet [27, 15] used a large part bank with sparse linear combination.  ...  The first is that one object can activate multiple overlapping hypotheses to pass through the whole cascade, while only one hypothesis with the highest score is useful for detection.  ... 
doi:10.1109/cvpr.2014.320 dblp:conf/cvpr/YanLWL14 fatcat:czuimg6sgfhl3mpcq3b2qknquy

Synapse classification and localization in Electron Micrographs

Vignesh Jagadeesh, James Anderson, Bryan Jones, Robert Marc, Steven Fisher, B.S. Manjunath
2014 Pattern Recognition Letters  
Algorithms such as the DPM and Sparselet have been shown to perform extremely well on such challenges.  ...  Strategies for efficiently utilizing an annotator's time through active learning is part of future work.  ... 
doi:10.1016/j.patrec.2013.06.001 fatcat:2xd3dtpjmfh75bt4edvwzg5z7q

Dynamical neural networks: Modeling low-level vision at short latencies

L. Perrinet
2007 The European Physical Journal Special Topics  
Using natural scenes, this algorithm provides a model of V1 which exhibit prototypical properties of neural activities in that area.  ...  (Right) Figure 25 : 25 Efficiency of the Sparselet Analysis. We compared here the sparselet analysis scheme at different levels of over-completeness with classical methods.  ...  In fact, a lower threshold would give more rapid results but at the same, the information needed to reach that threshold may not be enough to discriminate between different features.  ... 
doi:10.1140/epjst/e2007-00061-7 fatcat:ljct5ea6bnekrm2zolkx5rx3ky

Remote Sensing Image Scene Classification Meets Deep Learning: Challenges, Methods, Benchmarks, and Opportunities [article]

Gong Cheng, Xingxing Xie, Junwei Han, Lei Guo, Gui-Song Xia
2020 arXiv   pre-print
In general, a non-linear activation function f is performed after convolution operation.  ...  Cheng et al. [89] used the single-hidden-layer neural network and autoencoder for training more effective sparselets [90] to achieve efficient scene classification and object detection.  ... 
arXiv:2005.01094v1 fatcat:qz3at3gyvrbtzkluumalvpqb64

Efficient Evaluation of SVM Classifiers Using Error Space Encoding

Nisarg Raval, Rashmi Tonge, C.V. Jawahar
2014 2014 22nd International Conference on Pattern Recognition  
Instead of PCA, other methods like linear discriminant analysis (LDA) [5] can also be used to achieve the same goal.  ...  [15] proposed sparselet model based on sparse coding The idea is to represent a new query (SVM hyperplane) in terms of previously evaluated queries (w q = p i=1 α i w i ).  ... 
doi:10.1109/icpr.2014.755 dblp:conf/icpr/RavalTJ14 fatcat:unoldy7ewjfv5ewhh36nnc3jem

Part level transfer regularization for enhancing exemplar SVMs

Yusuf Aytar, Andrew Zisserman
2015 Computer Vision and Image Understanding  
Assuming that a positive value of α i represents the activation of the part u i and α i = 0 indicates that u i is not activated, the pairwise statistics can be captured through correlations between α i  ...  In [22, 42] HOG based templates are represented as a sparse reconstruction of shared parts (sparselets) which enable very fast evaluation of multiple detectors. Dean et al.  ... 
doi:10.1016/j.cviu.2015.04.004 fatcat:ydqjvoehwjfflflb4rikbr24hm

Transferring Pre-Trained Deep CNNs for Remote Scene Classification with General Features Learned from Linear PCA Network

Jie Wang, Chang Luo, Hanqiao Huang, Huizhen Zhao, Shiqiang Wang
2017 Remote Sensing  
Moreover, CaffeNet further places the non-linear activation functions after pooling layers [17] .  ...  Moreover, like wings of airplane, these parts are more discriminative with less blurs.  ... 
doi:10.3390/rs9030225 fatcat:ghpzuij5uzavhh7aaxbqfkxlra