Unsupervised Part-Based Weighting Aggregation of Deep Convolutional Features for Image Retrieval
Authors: Jian Xu, Cunzhao Shi, Chengzuo Qi, Chunheng Wang, Baihua Xiao
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct comprehensive experiments on four standard datasets and show that our unsupervised PWA outperforms the state-of-the-art unsupervised and supervised aggregation methods. |
| Researcher Affiliation | Academia | Jian Xu, Cunzhao Shi, Chengzuo Qi, Chunheng Wang, Baihua Xiao State Key Laboratory of Management and Control for Complex Systems, Institute of Automation, Chinese Academy of Sciences(CASIA) University of Chinese Academy of Sciences {xujian2015, cunzhao.shi, qichengzuo2013, chunheng.wang, baihua.xiao}@ia.ac.cn |
| Pseudocode | No | The paper describes the method mathematically and textually, but does not include any explicit pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide an explicit statement about the release of its source code or a link to a code repository for the methodology described. |
| Open Datasets | Yes | We evaluate the performance of PWA and other aggregation algorithms on four standard datasets (Oxford5k, Paris6k, Oxford105k and Paris106k) for image retrieval. Oxford5k (Philbin et al. 2007) and Paris6k (Philbin et al. 2008) datasets contain photographs collected from Flickr associated with Oxford and Paris landmarks respectively. |
| Dataset Splits | Yes | For fair comparison with the related retrieval methods, we learn the PCA and whitening parameters on Oxford5k when testing on Paris6k and vice versa. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for the experiments, such as GPU models, CPU types, or memory specifications. |
| Software Dependencies | No | Caffe (Jia et al. 2014) package for CNNs is used. [...] We extract deep convolutional features using the pre-trained VGG16 (Simonyan and Zisserman 2015) and fine-tuned Res Net101 from the work (Gordo et al. 2016b). (Does not specify version for Caffe or other libraries/frameworks) |
| Experiment Setup | Yes | We extract deep convolutional features using the pre-trained VGG16 (Simonyan and Zisserman 2015) and fine-tuned Res Net101 from the work (Gordo et al. 2016b). In the experiments, Caffe (Jia et al. 2014) package for CNNs is used. For VGG16 model, we extract convolutional feature maps from the pool5 layer and the number of channels is C=512. For Res Net-101 model, we extract convolutional feature maps from the res5c relu layer and the number of channels is C=2048. Regarding image size, we keep the original size of the images except for the very large images which are resized to the half size. The parameters for power normalization and power-scaling are set as α = 2 and β = 2, throughout our experiments. |