Explanation-based Data Augmentation for Image Classification
Authors: Sandareka Wickramanayake, Wynne Hsu, Mong Li Lee
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiment results demonstrate that the proposed approach improves the accuracy of classifiers compared to state-of-the-art augmentation strategies. We carry out experiments to show that BRACE can effectively improve the accuracy of general image classification tasks as well as fine-grained image classification. |
| Researcher Affiliation | Academia | Sandareka Wickramanayake Mong Li Lee Wynne Hsu School of Computing National University of Singapore {sandaw, leeml, whsu}@comp.nus.edu.sg |
| Pseudocode | Yes | Algorithm 1: BRACE input :Classification model M, Original training dataset D, Set of class labels C output :Fine-tuned weights θ |
| Open Source Code | No | The paper states 'All the codes are implemented using Py Torch', but does not provide any link or explicit statement about the availability of its own source code for the described methodology. |
| Open Datasets | Yes | Caltech UCSD Birds (CUB) [27]. This dataset has 11,788 images of birds belonging to 200 classes. The dataset is divided into a train set of 3994 images, a validation set of 2000 images, and a test set of 5794 images. CUB-Families [29]. Tiny Image Net. |
| Dataset Splits | Yes | Caltech UCSD Birds (CUB) [27]. This dataset has 11,788 images of birds belonging to 200 classes. The dataset is divided into a train set of 3994 images, a validation set of 2000 images, and a test set of 5794 images. The resultant dataset contains 4585 training images, 2343 validation images, and 4860 test images. In the standard dataset split, each class has 500 training images (400 for training the model and 100 for tuning the hyper-parameters) and 50 validation images. |
| Hardware Specification | Yes | All the codes are implemented using Py Torch, and the experiments are run on NVIDIA Tesla V100 GPUs. |
| Software Dependencies | No | The paper mentions 'Py Torch' but does not specify a version number or other software dependencies with their respective versions. |
| Experiment Setup | No | The paper states 'For each dataset, we set the number of nodes in the concept layer to be the same as the number of extracted word phrases and train CCNN using the same hyper-parameters settings in [17].' This refers to hyper-parameters being set as in a cited paper, but does not explicitly provide the specific values or training configurations within the main text of this paper. |