Fine-Grained Fashion Similarity Learning by Attribute-Specific Embedding Network
Authors: Zhe Ma, Jianfeng Dong, Zhongzi Long, Yao Zhang, Yuan He, Hui Xue, Shouling Ji11741-11748
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on four fashion-related datasets show the effectiveness of ASEN for fine-grained fashion similarity learning and its potential for fashion reranking. |
| Researcher Affiliation | Collaboration | Zhe Ma,1 Jianfeng Dong,2,3 Zhongzi Long,1 Yao Zhang,1 Yuan He,4 Hui Xue,4 Shouling Ji1,3 1Zhejiang University, 2Zhejiang Gongshang Unversity, 3Alibaba-Zhejiang University Joint Institute of Frontier Technologies, 4Alibaba Group |
| Pseudocode | No | The paper describes its proposed method using mathematical formulas and text, but does not include structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code and data are available at https://github.com/Maryeon/asen. |
| Open Datasets | Yes | As there are no existing datasets for attribute-specific fashion retrieval, we reconstruct three fashion datasets with attribute annotations to fit the task, i.e., Fashion AI (Zou et al. 2019), DARN (Huang et al. 2015) and Deep Fashion(Liu et al. 2016). For triplet relation prediction, we utilize Zappos50k (Yu and Grauman 2014). |
| Dataset Splits | Yes | We randomly split images into three sets by 8:1:1, which is 144k / 18k / 18k images for training / validation / test. Specifically, we use 70% / 10% / 20% images for training / validation / test. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU models, CPU types, or cloud computing instance specifications used for running its experiments. |
| Software Dependencies | No | The paper mentions general software components like 'CNN model pre-trained on Image Net', 'Res Net', 'Re LU function', 'sigmoid function', and 't-SNE', but does not provide specific version numbers for any libraries, frameworks, or solvers used in the experiments. |
| Experiment Setup | Yes | Concretely, we first construct a set of triplets T = {(I, I+, I |a)}, where I+ and I indicate images relevant and irrelevant with respect to image I in terms of attribute a. Given a triplet of {(I, I+, I |a)}, triplet ranking loss is defined as L(I, I+, I |a) = max{0, m s(I, I+|a) + s(I, I |a)}, where m represents the margin, empirically set to be 0.2. |