Zero-Shot Sketch-Based Image Retrieval via Graph Convolution Network

Authors: Zhaolong Zhang, Yuejie Zhang, Rui Feng, Tao Zhang, Weiguo Fan12943-12950

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our model gets a good performance on the challenging Sketchy and TU-Berlin datasets.
Researcher Affiliation Academia 1School of Computer Science, Shanghai Key Laboratory of Intelligent Information Processing, Fudan University, Shanghai, P.R. China 200433 {18210240044, yjzhang, fengrui}@fudan.edu.cn 2School of Information Management and Engineering, Shanghai Key Laboratory of Financial Information Technology, Shanghai University of Finance and Economics, Shanghai, P.R. China 200433 taozhang@mail.shufe.edu.cn 3Department of Business Analytics, Tippie College of Business, University of Iowa, Iowa City, Iowa, USA, 52242 weiguo-fan@uiowa.edu
Pseudocode No The paper describes the model architecture and mathematical formulations but does not include structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide a specific link or explicit statement about the public availability of the source code for the described methodology.
Open Datasets Yes We evaluate our proposed model on two popular SBIR datasets, i.e., Sketchy (Sangkloy et al. 2016) and TU-Berlin (Eitz, Hays, and Alexa 2012) by conducting extensive experiments.
Dataset Splits Yes We follow the setting in Yelamarthi et al. s work (2018) and split the dataset into 104 categories for training and 21 categories for testing, making sure that the testing categories do not appear in the 1,000 categories of Image Net. TU-Berlin-Extended... Following Shen et al. (2018), we randomly choose 30 categories that contain at least 400 images for testing and the rest 220 categories for training.
Hardware Specification Yes Our proposed Sketch GCN model is implemented with the popular deep learning toolbox Pytorch and trained on 4 TITAN Xp graphics cards.
Software Dependencies No The paper mentions implementation with 'Pytorch' but does not specify its version number or any other software dependencies with specific version numbers.
Experiment Setup Yes The model is optimized by the Adam algorithm with β1 = 0.9 and β2 = 0.999 across all the datasets, and the learning rates for the three part networks are set as lr1 = 0.00001, lr2 = lr3 = 0.0001 respectively. The weights of the loss are set as λ1 = 1, λ2 = 10, λ3 = 0.1.