Attention-Based Multi-Context Guiding for Few-Shot Semantic Segmentation
Authors: Tao Hu, Pengwan Yang, Chiliang Zhang, Gang Yu, Yadong Mu, Cees G. M. Snoek8441-8448
AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our architecture obtains state-of-the-art on unseen classes in a variant of PASCAL VOC12 dataset and performs favorably against previous work with large gains of 1.1%, 1.4% measured in m Io U in the 1-shot and 5-shot setting. Experimental Result Training details We implement our code based on the tensorflow framework (Abadi et al. 2016). Specially, a scaffold framework named tensorpack (Wu and others 2016) is used for quickly setting up our experiment. |
| Researcher Affiliation | Collaboration | Tao Hu,1,2 Pengwan Yang,2 Chiliang Zhang,3 Gang Yu,4 Yadong Mu,2 Cees G. M. Snoek1 1University of Amsterdam, 2Peking University, 3Tsinghua University, 4Megvii Inc. (Face++) |
| Pseudocode | No | The paper includes network diagrams and mathematical formulations, but no explicitly labeled 'Pseudocode' or 'Algorithm' sections, nor any structured code-like procedures. |
| Open Source Code | No | The paper does not provide concrete access to its source code, such as a specific repository link or an explicit code release statement. It only mentions the frameworks used: 'We implement our code based on the tensorflow framework (Abadi et al. 2016). Specially, a scaffold framework named tensorpack (Wu and others 2016) is used for quickly setting up our experiment.' |
| Open Datasets | Yes | We utilize dataset PASCAL-5i (Shaban et al. 2017) to conduct our experiment. This dataset is originated from PASCAL VOC12 (Everingham et al.) and extended annotations from SDS (Hariharan et al.). |
| Dataset Splits | Yes | Dataset: We utilize dataset PASCAL-5i (Shaban et al. 2017) to conduct our experiment. This dataset is originated from PASCAL VOC12 (Everingham et al.) and extended annotations from SDS (Hariharan et al.). The set of 20 classes in PASCAL VOC12 is divided into four sub-datasets as indicated in Table 2. Three sub-datasets are used as the training label-set Ltrain, the left one sub-dataset is utilized for test label-set Ltest. The training set Dtrain is composed of all image-mask pairs from PASCAL VOC12 and SDS training sets that include at least one pixel in the segmentation mask from the label-set Ltrain. The test set Dtest is from PASCAL VOC12 and SDS validation sets, and the processing procedure for test set Dtest is similar with training set Dtrain. Table 2: PASCAL-i5 group information. The top table displays 4 groups of label and their semantic classes. The bottom table shows 4 sub-datasets and their training, validation components. |
| Hardware Specification | Yes | All our models are trained by Stochastic Gradient Descent(SGD) (Bottou 2010) solver with learning rate=1e-4, momentum=0.99 on one Nvidia Titan XP GPU. |
| Software Dependencies | No | The paper mentions 'tensorflow framework (Abadi et al. 2016)' and 'tensorpack (Wu and others 2016)' but does not provide specific version numbers for these or any other software dependencies. |
| Experiment Setup | Yes | All our models are trained by Stochastic Gradient Descent(SGD) (Bottou 2010) solver with learning rate=1e-4, momentum=0.99 on one Nvidia Titan XP GPU. To fully fill GPU memory, we set the batch size 12. The weights of the support branch and the query branch are initialized with Image Net (Deng et al. 2009) pre-trained weights. For the weight initialization of A-MCG module, Xavier initialization (Glorot and Bengio 2010) is adopted. All the images in the support and query branch are resized to 320 320. |