Learning Intact Features by Erasing-Inpainting for Few-shot Classification
Authors: Junjie Li, Zilei Wang, Xiaoming Hu8401-8409
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The extensive experiments on two widely used benchmarks well demonstrates the effectiveness of our proposed method, which can consistently get considerable performance gains for different baseline methods. |
| Researcher Affiliation | Academia | Junjie Li, Zilei Wang*, Xiaoming Hu Department of Automation, University of Science and Technology of China hnljj@mail.ustc.edu.cn, zlwang@ustc.edu.cn, cjdc@mail.ustc.edu.cn |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide an explicit statement or link for open-source code availability for the described methodology. |
| Open Datasets | Yes | We use mini Image Net (Vinyals et al. 2016) and tired Image Net (Ren et al. 2018) to evaluate the performance of different methods. |
| Dataset Splits | Yes | According to the standard split (Vinyals et al. 2016), the 100 classes are divided into 64 training classes, 16 validation classes, and 20 test classes, respectively. The tired Image Net is a larger dataset, which totally consists of 34 categories (608 classes), and have 779, 165 images for training, validation, and test. The 34 categories (608 classes) are divided into 20 training categories (351 classes), 6 validation categories (97 classes), and 8 test categories (160 classes). |
| Hardware Specification | Yes | Py Torch (Paszke et al. 2017) and NVIDIA 1080Ti GPUs are used throughout our experiments. |
| Software Dependencies | No | The paper mentions 'Py Torch' but does not specify a version number. |
| Experiment Setup | Yes | For mini Image Net, the model is trained for 110 epochs, each of which contains 1, 200 episodes. For tired Image Net, the model is trained for 100 epochs, each of which contains 13, 980 episodes. Particularly, for each training episode, we randomly sample 30 query samples. In the test stage, each episode consists of 75 query samples. |