Adversarial Feature Augmentation and Normalization for Visual Recognition
Authors: Tianlong Chen, Yu Cheng, Zhe Gan, Jianfeng Wang, Lijuan Wang, Jingjing Liu, Zhangyang Wang
TMLR 2022 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments show that A-FAN yields consistent generalization improvement over strong baselines across various datasets for classification, detection and segmentation tasks, such as CIFAR-10, CIFAR-100, Image Net, Pascal VOC2007, Pascal VOC2012, COCO2017, and Cityspaces. Comprehensive ablation studies and detailed analyses also demonstrate that adding perturbations to specific modules and layers of classification/detection/segmentation backbones yields optimal performance. |
| Researcher Affiliation | Collaboration | Tianlong Chen EMAIL University of Texas at Austin Yu Cheng EMAIL Microsoft Research Zhe Gan EMAIL Microsoft Cloud & AI Jianfeng Wang EMAIL Microsoft Cloud & AI Lijuan Wang EMAIL Microsoft Cloud & AI Jingjing Liu EMAIL Tsinghua University Zhangyang Wang EMAIL University of Texas at Austin |
| Pseudocode | Yes | Algorithm 1 Adversarial Feature Augmentation and Normalization (A-FAN). |
| Open Source Code | Yes | Codes and pre-trained models are available in: https://github.com/VITA-Group/CV_A-FAN. |
| Open Datasets | Yes | Extensive experiments show that A-FAN yields consistent generalization improvement over strong baselines across various datasets for classification, detection and segmentation tasks, such as CIFAR-10, CIFAR-100, Image Net, Pascal VOC2007, Pascal VOC2012, COCO2017, and Cityspaces. Comprehensive ablation studies and detailed analyses also demonstrate that adding perturbations to specific modules and layers of classification/detection/segmentation backbones yields optimal performance. |
| Dataset Splits | Yes | In our experiments, the original training datasets are randomly split into 90% training and 10% validation. The early stopping technique is applied to find the top-performing checkpoints on the validation set. Then, the selected checkpoints are evaluated on the test set to report the performance. (from Section 4). Also, in Pascal VOC2007, we use the train and validation sets for training, and evaluate on test set; in COCO2017, we train models on the train set and evaluateon the validation set. (from Section 5) |
| Hardware Specification | Yes | CIFAR-10 models are trained on a single GTX1080 Ti GPU. Image Net (batch size 256) experiments are conducted on 2 Quadro RTX 6000 GPUs with 24G 2 memory in total. |
| Software Dependencies | No | No specific software versions (e.g., PyTorch 1.x, CUDA x.x, Python x.x) are explicitly mentioned in the paper, only general software usage like 'Pytorch repository'. |
| Experiment Setup | Yes | For network training on CIFAR-10 and CIFAR-100, we adopt an SGD optimizer with a momentum of 0.9, weight decay of 5 10 4, and batch size of 128 for 200 epochs. The learning rate starts from 0.1 and decays to one-tenth at 50-th and 150-th epochs. We also perform a linear learning rate warm-up in the first 200 iterations. For Image Net experiments, following the official setting in Pytorch repository,2 we train deep networks for 90 epochs with a batch size of 512, and the learning rate decay at 30-th and 60-th epoch. The SGD optimizer is adopted with a momentum of 0.9 and a weight decay of 1 10 4. (from Section 4). Also, Table 5 titled "Details of training and evaluation" provides specific hyperparameters for different datasets and tasks. |