Order-Free RNN With Visual Attention for Multi-Label Classification

Authors: Shang-Fu Chen, Yi-Chen Chen, Chih-Kuan Yeh, Yu-Chiang Wang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments on NUS-WISE and MS-COCO datasets confirm the design of our network and its effectiveness in solving multi-label classification problems.
Researcher Affiliation Academia Shang-Fu Chen,1 Yi-Chen Chen,2 Chih-Kuan Yeh,3 Yu-Chiang Frank Wang1,2 1Department of Electrical Engineering, National Taiwan University, Taipei, Taiwan 2Graduate Institute of Communication Engineering, National Taiwan University, Taipei, Taiwan 3Machine Learning Department, Carnegie Mellon University, Pittsburgh, USA
Pseudocode Yes Algorithm 1: Training of Our Proposed Model
Open Source Code No The paper does not provide any explicit statements about making its source code publicly available, nor does it include a link to a code repository.
Open Datasets Yes Our experiments on NUS-WISE and MS-COCO datasets confirm the design of our network and its effectiveness in solving multi-label classification problems. [...] NUS-WIDE is a web image dataset which includes 269,648 images [...] The training set consists of 82,783 images with up to 80 annotated object labels.
Dataset Splits No For NUS-WIDE, the paper states "150,000 images are considered for training, and the rest for testing" (a train/test split). For MS-COCO, it notes "The test set of this experiment utilizes the validation set of MS-COCO (40,504 images)" indicating the MS-COCO validation set was used as their test set. It mentions "We perform validation on the stopping threshold for beam search" but this refers to a hyperparameter tuning process, not a dedicated dataset split for validation.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments (e.g., GPU models, CPU types, or memory specifications).
Software Dependencies No The paper mentions using a "Res Net152 network" and an "Adam optimizer" but does not specify any software names with version numbers for libraries or frameworks used (e.g., TensorFlow, PyTorch, scikit-learn, Python version).
Experiment Setup Yes We employ the Adam optimizer with the learning rate at 0.0003, and the dropout rate at 0.8 for updating fpred.