Efficient Image Retrieval via Decoupling Diffusion into Online and Offline Processing

Authors: Fan Yang, Ryota Hinami, Yusuke Matsui, Steven Ly, Shin’ichi Satoh9087-9094

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The results on Oxford5k and Oxford105k datasets are shown in Fig. 4. Table 1 compares our method with other competitive methods that use global and regional features.
Researcher Affiliation Academia 1The University of Tokyo, Japan 2National Institute of Informatics, Japan 3University of Southern California, USA
Pseudocode Yes Algorithm 1 Online search
Open Source Code Yes The source code to replicate our experiments is available at https://github.com/fyang93/diffusion.
Open Datasets Yes We use the Oxford Buildings (Philbin et al. 2007) and Paris (Philbin et al. 2008) datasets in our experiments.
Dataset Splits No The paper mentions using Oxford and Paris datasets, and adding distractors to create larger datasets, but does not specify the exact training, validation, or testing splits used for its models.
Hardware Specification Yes For the efficiency evaluation, we use a single core of Intel Xeon 2.80GHz CPU. using a single GPU per image.
Software Dependencies No The paper mentions using the FAISS toolkit and other standard libraries, but does not provide specific version numbers for these software dependencies (e.g., 'FAISS 1.x' or 'Python 3.x').
Experiment Setup Yes We use the same graph construction parameters as in the previous work (Iscen et al. 2017). In particular, the parameter α to build Lα is set to 0.99. For global features, 50 nearest neighbors of each database element are used for graph construction, and the initial state vector contains the similarities between the query and its 10 nearest neighbors. While for regional features, the corresponding numbers of nearest neighbors are set to 200, 200 respectively.