Deep Semantic Structural Constraints for Zero-Shot Learning

Authors: Yan Li, Zhen Jia, Junge Zhang, Kaiqi Huang, Tieniu Tan

AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments Datasets We evaluate the DSSC model and compare with existing state-of-the-art approaches on three standard zero-shot learning benchmark datasets. Table 1 summarizes their statistics. Animals with Attributes (Aw A) (Lampert, Nickisch, and Harmeling 2014) includes 30,475 images from 50 animals categories. We adopt the class-level continuous 85-dim attributes as the semantic representations and use the standard 40/10 zero-shot split.
Researcher Affiliation Academia Yan Li, 1,2 Zhen Jia, 1,2 Junge Zhang,1,2 Kaiqi Huang,1,2,3 Tieniu Tan1,2,3 1 CRIPAC & NLPR, Institute of Automation, Chinese Academy of Sciences 2 University of Chinese Academy of Sciences 3 CAS Center for Excellence in Brain Science and Intelligence Technology
Pseudocode No The paper does not contain any clearly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not provide an explicit statement about releasing source code or a link to a code repository.
Open Datasets Yes Animals with Attributes (Aw A) (Lampert, Nickisch, and Harmeling 2014) includes 30,475 images from 50 animals categories. Caltech-UCSD Birds 200-2011 (CUB) (Wah et al. 2011) is a fine-grained bird dataset with 200 different species of birds and 11,788 images. SUN-Attribute (SUN) (Patterson et al. 2014) contains 14,340 images coming from 717 fine-grained scenes.
Dataset Splits No The paper specifies training and test class splits for datasets (e.g., '40/10 zero-shot split' for Aw A, '150/50 zero-shot split' for CUB, '645 classes for training and 72 classes for test' for SUN), but it does not explicitly define a separate validation dataset split.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments, such as GPU or CPU models.
Software Dependencies No The paper mentions models like 'Goog Le Net' and 'Res Net-101' and datasets like 'Image Net' and 'MIT Places dataset', but does not list specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes LIF Sc = max(0, m IF Sc + d(φ(xi), φ(xk)) d(φ(xi), φ(xj))) where ... m IF Sc is the margin of the IFSc and is set to 1.0 for all experiments. LDSSC = Lsoftmax + λLIF Sc + βLSESc where λ and β are trivially set to 1.0 for all the experiments. During the training stage, all images are resized to 256 256 pixels.