GuidedMix-Net: Semi-supervised Semantic Segmentation by Using Labeled Images as Reference
Authors: Peng Tu, Yawen Huang, Feng Zheng, Zhenyu He, Liujuan Cao, Ling Shao2379-2387
AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on PASCAL VOC 2012, and Cityscapes demonstrate the effectiveness of our Guided Mix-Net, which achieves competitive segmentation accuracy and significantly improves the m Io U over 7% compared to previous approaches. |
| Researcher Affiliation | Collaboration | Peng Tu*1, 2, Yawen Huang*3, Zheng Feng 1, Zhenyu He4, Liujuan Cao5, Ling Shao6 1Southern University of Science and Technolog, Shenzhen, China 2Shenzhen Microbt Electronics Technology Co., Ltd, China 3Tencent Jarvis Lab, Shenzhen, China 4Harbin Institute of Technology, Shenzhen, China 5Xiamen University, Xiamen, China 6National Center for Artificial Intelligence, Saudi Data and Artificial Intelligence Authority, Riyadh, Saudi Arabia |
| Pseudocode | No | The paper does not contain any pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any specific links or explicit statements about the release of source code for the described methodology. |
| Open Datasets | Yes | PASCAL VOC 2012. This dataset is widely used for semantic segmentation and object detection. It consists of 21 classes including background. We use 1,464 training images and 1,449 validation images from the original PASCAL dataset, and also leverage the augmented annotation dataset (involving 9,118 images) (Hariharan et al. 2011). Cityscapes. This dataset provides different driving scenes distributed in 19 classes, with 2,975, 500, 1,525 densely annotated images for training, validation and testing. |
| Dataset Splits | Yes | We use 1,464 training images and 1,449 validation images from the original PASCAL dataset, and also leverage the augmented annotation dataset (involving 9,118 images) (Hariharan et al. 2011). Cityscapes. This dataset provides different driving scenes distributed in 19 classes, with 2,975, 500, 1,525 densely annotated images for training, validation and testing. |
| Hardware Specification | Yes | We conduct all our experiments on a Tesla V-100s GPU. |
| Software Dependencies | No | The paper mentions using Res Net and U-Net architectures but does not specify any software names with version numbers (e.g., Python, PyTorch, TensorFlow versions). |
| Experiment Setup | Yes | Similar to (Chen et al. 2017), we use a poly learning rate policy, where the base learning rate is multiplied by ((1 iter maxier)power) and power = 0.9. Our segmentation network is optimized using the stochastic gradient descent (SGD) optimizer with a base learning rate of 1e-3, momentum of 0.9 and a weight decay of 1e-4. The model is trained over 40,000 iterations for all datasets, and the batch-size is set to 12 for PASCAL VOC 2012, and 8 for Cityscapes and PASCAL Context. |