TransNFCM: Translation-Based Neural Fashion Compatibility Modeling

Authors: Xun Yang, Yunshan Ma, Lizi Liao, Meng Wang, Tat-Seng Chua403-410

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments demonstrate the effectiveness of Trans NFCM over the state-of-the-arts on two real-world datasets.
Researcher Affiliation Academia 1School of Computing, National University of Singapore, Singapore 2School of Computing and Information Engineering, Hefei University of Technology, China
Pseudocode No The paper describes the model and optimization process mathematically and textually, but does not include any pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any statement about releasing source code or a link to a code repository.
Open Datasets Yes We conduct experiments on two public fashion compatibility datasets, both crawled from the fashion social commerce site Polyvore (www.polyvore.com)... Fashion VC (Song et al. 2017). It was collected for topbottom recommendation... Polyvore Maryland (Han et al. 2017). It was released by (Han et al. 2017) for outfit compatibility modeling.
Dataset Splits Yes Fashion VC (Song et al. 2017). It was collected for topbottom recommendation, consisting of 14,871 tops and 13,663 bottoms, split into 80% for training, 10% for validation, and 10% for testing. ... We extract and resplit the co-occurring item pairs randomly in the same setting with Fashion VC: 80% for training, 10% for testing, and 10% for validation.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments (e.g., CPU, GPU models, memory).
Software Dependencies No The paper mentions software like 'Pytorch', 'Alex Net', and 'text CNN architecture', but does not specify their version numbers. It also mentions 'word2vec vector' without specifying a version.
Experiment Setup Yes We set the overall learning rate η = 0.001, and drop it to η = η/10 every 10 epochs. The learning rate of the pretrained Alex Net in V-Encoder is set to η = η/10 for fine-tuning. The margin γ is set to 1 following the setting of (Bordes et al. 2013). We use 128 5-tuples in a minibatch. Both the dimensions of visual embedding vectors and textual embedding vectors are set to d = 128. Dropout is used in both visual and textual encoders.