Dictionary Learning Inspired Deep Network for Scene Recognition

Authors: Yang Liu, Qingchao Chen, Wei Chen, Ian Wassell

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our proposed approach is evaluated using various scene datasets and shows superior performance to many stateof-the-art approaches.
Researcher Affiliation Academia 1Department of Computer Science and Technology, University of Cambridge, UK 2Department of Security and Crime Science, University College London, UK 3State Key Laboratory of Rail Traffic Control and Safety, Beijing Jiaotong University, China
Pseudocode No The paper describes the operations of the non-linear DLL using mathematical equations (6-9) and data flow diagrams (Figure 2), but does not present a formal pseudocode or algorithm block with structured steps formatted like code.
Open Source Code No The paper does not provide any explicit statement about releasing open-source code or a link to a code repository for the methodology described.
Open Datasets Yes We employ these three widely used scene recognition datasets in our experiments. We use the average accuracy to evaluate the recognition performance. The parameters in the objective function (3) are determined by 5-fold cross-validation for different datasets as listed in Table. 1. We follow the same training-test partition used in (Liu et al. 2014) (Xie et al. 2017). 15 Scene includes 100 images per class for training and the rest for testing. MIT Indoor 67 includes 80 images of each category for training and 20 images for testing. SUN 397 includes multiple train/test splits, with 50 images per class in the testing set.
Dataset Splits Yes The parameters in the objective function (3) are determined by 5-fold cross-validation for different datasets as listed in Table. 1. We follow the same training-test partition used in (Liu et al. 2014) (Xie et al. 2017). 15 Scene includes 100 images per class for training and the rest for testing. MIT Indoor 67 includes 80 images of each category for training and 20 images for testing. SUN 397 includes multiple train/test splits, with 50 images per class in the testing set.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper does not provide specific software dependency details, such as library names with version numbers, used to replicate the experiment.
Experiment Setup Yes The parameters in the objective function (3) are determined by 5-fold cross-validation for different datasets as listed in Table. 1. We used K = 2 recurrent units within each DLL throughout the paper unless otherwise specified. We only replace last two FCLs by an equivalent number of the proposed non-linear DLLs. The parameters of the dictionary are initialized using a truncated normal distribution, with the number of dictionary atoms per class varying from 5 to 70.