Filters








1,091 Hits in 2.4 sec

Sparse transfer learning for interactive video search reranking

Xinmei Tian, Dacheng Tao, Yong Rui
2012 ACM Transactions on Multimedia Computing, Communications, and Applications (TOMCCAP)  
However, existing reranking algorithms can only achieve limited improvement because of the well-known semantic gap between low level visual features and high level semantic concepts.  ...  Technically, it a) considers the pair-wise discriminative information to maximally separate labeled query relevant samples from labeled query irrelevant ones, b) achieves a sparse representation for the  ...  The dimension reduction, which aims to find a compact representation for the samples in a low dimensional subspace, is a suitable candidate.  ... 
doi:10.1145/2240136.2240139 fatcat:qvnayxhxnfcwtgu6di5j2suwna

Sparse Transfer Learning for Interactive Video Search Reranking [article]

Xinmei Tian and Dacheng Tao and Yong Rui
2011 arXiv   pre-print
However, existing reranking algorithms can only achieve limited improvement because of the well-known semantic gap between low level visual features and high level semantic concepts.  ...  Technically, it a) considers the pair-wise discriminative information to maximally separate labeled query relevant samples from labeled query irrelevant ones, b) achieves a sparse representation for the  ...  N w          and the corresponding low Sparse Transfer Learning for Interactive Video Search Reranking • 20: 7 dimensional feature matrix is given by 11 [ , ,  ... 
arXiv:1103.2756v2 fatcat:ixdqa7ltkbhfdccz5vgihz7nqe

Attribute Based Image Search Re-Ranking

2015 International Journal of Science and Research (IJSR)  
Image search reranking is an effective approach to refine the text-based image search result.  ...  In this paper, reranking methods are suggested address this problem in scalable fashion.  ...  Finally, obtain a 9,744-dimensional feature for each image, consisting of a 1, 392 × 6-dimensional feature from the grids and a 1,392-dimensional feature from the image.  ... 
doi:10.21275/v4i11.nov151336 fatcat:j6lepns43zf3xb3vnt6q245kba

Improving person re-identification by soft biometrics based reranking

Le An, Xiaojing Chen, Mehran Kafai, Songfan Yang, Bir Bhanu
2013 2013 Seventh International Conference on Distributed Smart Cameras (ICDSC)  
These distance scores are used for reranking the initially returned matches.  ...  During reranking, the soft biometric attributes are detected and attribute-based distance scores are calculated between pairs of images by using a regression model.  ...  Since the feature dimension (5 + 5 = 10 in our case) is significantly smaller than the number of samples, we use Radial Basis Function (RBF) kernel to map the lower dimensional features into higher dimensional  ... 
doi:10.1109/icdsc.2013.6778216 dblp:conf/icdsc/AnCKYB13 fatcat:jx3tnjplhratdc3h6t4472hskm

Segmentation-free word spotting with exemplar SVMs

Jon Almazán, Albert Gordo, Alicia Fornés, Ernest Valveny
2014 Pattern Recognition  
Then, we use a more discriminative representation based on Fisher Vector to rerank the best regions retrieved, and the most promising ones are used to expand the Exemplar SVM training set and improve the  ...  First, they can be improved in the choice of low level features. The features of [6, 7, 8, 14] produce sequence representations, which are usually slower to compare than fixed-length representations.  ...  Unfortunately, vector quantization is not possible when the dimensionality of the vectors is not trivially low: as noted in [19] , to encode a descriptor of 128 dimensions using only 0.5 bits per dimension  ... 
doi:10.1016/j.patcog.2014.06.005 fatcat:7r4a6krnubdujmjspw2g34onxi

Semi-supervised training for statistical word alignment

Alexander Fraser, Daniel Marcu
2006 Proceedings of the 21st International Conference on Computational Linguistics and the 44th annual meeting of the ACL - ACL '06  
We introduce a semi-supervised approach to training for statistical machine translation that alternates the traditional Expectation Maximization step that is applied on a large training corpus with a discriminative  ...  Discriminative Reranking with Improved Search.  ...  The low F-measure scores of the baselines motivate our work.  ... 
doi:10.3115/1220175.1220272 dblp:conf/acl/FraserM06 fatcat:62dmq3ptebcixpjmjjfx3kvq7u

Content-Based Video Search over 1 Million Videos with 1 Core in 1 Second

Shoou-I Yu, Lu Jiang, Zhongwen Xu, Yi Yang, Alexander G. Hauptmann
2015 Proceedings of the 5th ACM on International Conference on Multimedia Retrieval - ICMR '15  
accuracy trade-off of reranking.  ...  Directions investigated include exploring different low-level and semanticsbased features, testing different compression factors and approximations during video search, and understanding the time v.s.  ...  Effective low-level visual features include the Improved Dense Trajectories Fisher Vectors (IDT FV, 110592 dimensional) from [27, 4] .  ... 
doi:10.1145/2671188.2749398 dblp:conf/mir/YuJXYH15 fatcat:p2ios43a75anjjkuswkrrfuk44

Efficient Stacked Dependency Parsing by Forest Reranking

Katsuhiko Hayashi, Shuhei Kondo, Yuji Matsumoto
2013 Transactions of the Association for Computational Linguistics  
This paper proposes a discriminative forest reranking algorithm for dependency parsing that can be seen as a form of efficient stacked parsing.  ...  A dynamic programming shift-reduce parser produces a packed derivation forest which is then scored by a discriminative reranker, using the 1-best tree output by the shift-reduce parser as guide features  ...  The MERT algorithm is suited to tune low-dimensional parameters. The β was set to about 1.2 in case of local reranking, and to about 1.5 in case of non-local reranking.  ... 
doi:10.1162/tacl_a_00216 fatcat:nmtftcqitzgvbkfuisuazcsviy

Accessible image search

Meng Wang, Bo Liu, Xian-Sheng Hua
2009 Proceedings of the seventeen ACM international conference on Multimedia - MM '09  
Based on the measured accessibility scores, different reranking methods can be performed to prioritize the images with high accessibilities.  ...  Rasche et al. formulate the recoloring task as a dimensionality reduction problem, i.e., how to map the colors in a 3-dimensional space into a 2dimensional space that can be recognized by colorblind viewers  ...  Protanopes and deuteranopes have difficulty in discriminating red from green, whereas tritanopes have difficulty in discriminating blue from yellow.  ... 
doi:10.1145/1631272.1631314 dblp:conf/mm/WangLH09 fatcat:osshwy7fefckfmtzhr7alpbxeu

AliExpress Learning-To-Rank: Maximizing Online Model Performance without Going Online [article]

Guangda Huzhang, Zhen-Jia Pang, Yongqing Gao, Yawen Liu, Weijie Shen, Wen-Ji Zhou, Qing Da, An-Xiang Zeng, Han Yu, and Yang Yu, Zhi-Hua Zhou
2020 arXiv   pre-print
The framework consists of an evaluator that generalizes to evaluate recommendations involving the context, and a generator that maximizes the evaluator score by reinforcement learning, and a discriminator  ...  • We present the EG-Rerank and EG-Rerank+ as standard approaches of the evaluator-generator framework.  ...  EG-Rerank+ consistently improves the conversion rate by 2% over the fine-tuned industrial-level re-ranking model in online A/B tests, which is a significant improvement in a large-scale e-commerce platform  ... 
arXiv:2003.11941v5 fatcat:rwtqltj3xnfirj6r7dzrwi2wly

Rethink Training of BERT Rerankers in Multi-Stage Retrieval Pipeline [article]

Luyu Gao, Zhuyun Dai, Jamie Callan
2021 arXiv   pre-print
Rerankers fine-tuned from deep LM estimates candidate relevance based on rich contextualized matching signals.  ...  In this paper, we discover otherwise and that popular reranker cannot fully exploit the improved retrieval result.  ...  A discriminative reranker should be able to handle the top portion of retriever results and avoid relying on those confounding features.  ... 
arXiv:2101.08751v1 fatcat:vq6xunkgz5gcjlcgffktbvfquu

Immediate, Scalable Object Category Detection

Yusuf Aytar, Andrew Zisserman
2014 2014 IEEE Conference on Computer Vision and Pattern Recognition  
indexing; (ii) a sparse representation of a HOG classifier using a set of mid-level discriminative classifier patches; and (iii) a fast method for spatial reranking images on their detections.  ...  We evaluate the detection method on the standard PAS-CAL VOC 2007 dataset, together with a 100K image subset of ImageNet, and demonstrate near state of the art detection performance at low ranks whilst  ...  Again, the FS with reranking performance at low ranks is quite good -though not at the same level as for the DPM trained templates of table 2.  ... 
doi:10.1109/cvpr.2014.305 dblp:conf/cvpr/AytarZ14 fatcat:dzk5z5uisrgsbd6ls45oc5jkdq

Where to Focus: Query Adaptive Matching for Instance Retrieval Using Convolutional Feature Maps [article]

Jiewei Cao, Lingqiao Liu, Peng Wang, Zi Huang, Chunhua Shen, Heng Tao Shen
2016 arXiv   pre-print
In this article, we alleviate this drawback by proposing a novel reranking algorithm using CFMs to refine the retrieval result obtained by existing methods.  ...  Through extensive experiments, we show that our reranking approaches bring substantial performance improvement and by applying them we can outperform the state of the art on several instance retrieval  ...  dimensionality settings.  ... 
arXiv:1606.06811v1 fatcat:tm56q4sptrg3tlflydmhlrqely

Forest Reranking: Discriminative Parsing with Non-Local Features

Liang Huang
2008 Annual Meeting of the Association for Computational Linguistics  
We instead propose forest reranking, a method that reranks a packed forest of exponentially many parses.  ...  Since exact inference is intractable with non-local features, we present an approximate algorithm inspired by forest rescoring that makes discriminative training practical over the whole Treebank.  ...  This method can thus be viewed as a step towards the integration of discriminative reranking with traditional chart parsing.  ... 
dblp:conf/acl/Huang08 fatcat:kygyhok6njagjgdge5wrbq7fcq

Groupwise learning for ASR k-best list reranking in spoken language translation

Raymond W. M. Ng, Kashif Shah, Lucia Specia, Thomas Hain
2016 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)  
Groupwise learning is robust to sentences with different ASR-confidence, meaning that the confidence threshold heuristics in reranking are no longer needed.  ...  This technique is also complementary to linear discriminant analysis feature projection. Altogether a BLEU score improvement of 0.80 was achieved in an indomain English-to-French SLT task.  ...  LDA aims to find a projection of the feature vector to a low dimensional space subject to the Fisher criterion, and was shown to give an extra 0.04-0.11 BLEU score increase in previous SLT enhancement  ... 
doi:10.1109/icassp.2016.7472853 dblp:conf/icassp/NgSSH16 fatcat:aqrrg7tij5bxhf36kautkcqlha
« Previous Showing results 1 — 15 out of 1,091 results