Vision GNN: An Image is Worth Graph of Nodes
Authors: Kai Han, Yunhe Wang, Jianyuan Guo, Yehui Tang, Enhua Wu
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on image recognition and object detection tasks demonstrate the superiority of our Vi G architecture. |
| Researcher Affiliation | Collaboration | Kai Han1,2 Yunhe Wang2 Jianyuan Guo2 Yehui Tang2,3 Enhua Wu1,4 1State Key Lab of Computer Science, ISCAS & UCAS 2Huawei Noah s Ark Lab 3Peking University 4University of Macau {kai.han,yunhe.wang}@huawei.com, weh@ios.ac.cn |
| Pseudocode | No | The paper describes the model architecture and components in detail within the main text and through equations (Eq. 1-7), but it does not include any explicitly labeled 'Pseudocode' or 'Algorithm' block. |
| Open Source Code | Yes | The Py Torch code is available at https://github.com/huawei-noah/ Efficient-AI-Backbones and the Mind Spore code is available at https: //gitee.com/mindspore/models. |
| Open Datasets | Yes | In image classification task, the widely-used benchmark Image Net ILSVRC 2012 [43] is used in the following experiments. ... For the license of Image Net dataset, please refer to http://www.image-net.org/download. For object detection, we use COCO 2017 [34] dataset with 80 object categories. ... For the licenses of these datasets, please refer to https://cocodataset.org/#home. |
| Dataset Splits | Yes | Image Net has 1.2M training images and 50K validation images, which belong to 1000 categories. ... COCO 2017 contains 118K training images and 5K validation images. ... All the models are trained on COCO 2017 training set in 1 schedule and evaluated on validation set. |
| Hardware Specification | Yes | We implement the networks using Py Troch [41] and Mind Spore [23] and train all our models on 8 NVIDIA V100 GPUs. |
| Software Dependencies | No | The paper mentions using 'Py Troch [41]' and 'Mind Spore [23]' for implementation, and 'Adam W [37]' as an optimizer, and 'GELU [19]' as an activation function. However, it does not provide specific version numbers for any of these software dependencies. |
| Experiment Setup | Yes | Table 3: Training hyper-parameters for Image Net. (Pyramid) Vi G Ti S M B Epochs 300 Optimizer Adam W [37] Batch size 1024 Start learning rate (LR) 2e-3 Learning rate schedule Cosine Warmup epochs 20 Weight decay 0.05 Label smoothing [47] 0.1 Stochastic path [22] 0.1 0.1 0.1 0.3 Repeated augment [20] Rand Augment [6] Mixup prob. [73] 0.8 Cutmix prob. [72] 1.0 Random erasing prob. [74] 0.25 Exponential moving average 0.99996 |