Graph Knows Unknowns: Reformulate Zero-Shot Learning as Sample-Level Graph Recognition

Authors: Jingcai Guo, Song Guo, Qihua Zhou, Ziming Liu, Xiaocheng Lu, Fushuo Huo

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on the widely used benchmark datasets demonstrate that the proposed method can mitigate the domain bias problem and achieve competitive performance against other representative methods.
Researcher Affiliation Academia Jingcai Guo1,2, Song Guo1,2, Qihua Zhou1, Ziming Liu1, Xiaocheng Lu1,3, Fushuo Huo1 1Department of Computing, The Hong Kong Polytechnic University, Hong Kong SAR, China 2The Hong Kong Polytechnic University Shenzhen Research Institute, Shenzhen, China 3School of Computer Science, Northwestern Polytechnical University, Xi an, China
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide concrete access to source code for the methodology described.
Open Datasets Yes Following (Ji et al. 2018; Elhoseiny et al. 2017), we evaluate our method on two widely used fine-grained datasets including CUB-Birds (Wah et al. 2011) and NABirds (Van Horn et al. 2015).
Dataset Splits Yes For ZSL, 150 classes of bird images act as seen classes for training, and the remaining 50 classes are unseen classes. ... Among them, 323 classes are seen classes and the remaining 81 are unseen classes. Similarly, each image is also annotated with the required key-point location, while differently, the semantic descriptions of each class is a collected article from Wikipedia.
Hardware Specification Yes Our method is implemented by Pytorch and trained with NVIDIA RTX 3090 GPU.
Software Dependencies No The paper mentions 'Pytorch' but does not specify a version number for it or any other software dependency.
Experiment Setup Yes The GNNs consist of four graph convolution layers... a Res Net-34 with the classification layer removed is acted as an encoder... The cropped element size w and h are both set to 56 for CUB-Birds and 224 for NABirds... we control the threshold ε to retain = 50 and = 20 edges among the sample-level graph for CUB-Birds and NABirds, respectively.