Zero-Shot Ingredient Recognition by Multi-Relational Graph Convolutional Network

Authors: Jingjing Chen, Liangming Pan, Zhipeng Wei, Xiang Wang, Chong-Wah Ngo, Tat-Seng Chua10542-10550

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on both Chinese and Japanese food datasets are performed to demonstrate the superior performance of multi-relational GCN and shed light on zero-shot ingredients recognition.
Researcher Affiliation Academia 1Shanghai Key Lab of Intelligent Information Processing, School of Computer Science, Fudan University; 2NUS Graduate School for Integrative Sciences and Engineering, National University of Singapore; 3School of Computing, National University of Singapore; 4Jilin University; 5City University of Hong Kong
Pseudocode No The paper does not contain any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any explicit statement or link indicating that the source code for their methodology is publicly available.
Open Datasets Yes The performances are evaluated on two datasets, a Chinese food dataset VIREO Food 172 (Chen and Ngo 2016) and a Japanese food dataset UEC Food-100 (Matsuda and Yanai 2012).
Dataset Splits Yes VIREO Food-172 covers 172 most common Chinese dishes, being labeled with 353 ingredients. In total, this dataset contains 110,241 food images. We split 60% as the training set, 30% as the test set and the remaining for validation.
Hardware Specification No The paper mentions using ResNet-50 and VGG-16, which are model architectures, but does not specify any hardware details like GPU models, CPU types, or memory used for experiments.
Software Dependencies No The paper mentions using Adam optimizer, Leaky ReLU activation, and word2vec, ResNet-50, VGG-16 models, but does not provide specific version numbers for any of these software components or libraries.
Experiment Setup Yes For training the m RGCN, we adopt Adam (Kingma and Ba 2014) optimizer and the learning rate is set as 0.001. We train the m RGCN for 500 epochs for every experiment and report the performance of the model which attains the best result on validation set. Similar to (Wang, Ye, and Gupta 2018), we apply Leaky Re LU (Maas, Hannun, and Ng 2013) with the negative slope of 0.2 as the activation function and perform L2 normalization on the output of m RGCN to regularize the output into similar magnitudes. ... our m RGCN is composed of 4 convolutional layers with output channel numbers as 2048-1024-512-D.