Distilling Inter-Class Distance for Semantic Segmentation
Authors: Zhengbo Zhang, Chunluan Zhou, Zhigang Tu
IJCAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on three popular datasets: Cityscapes, Pascal VOC and ADE20K show that our method is helpful to improve the accuracy of semantic segmentation models and achieves the state-of-the-art performance. |
| Researcher Affiliation | Collaboration | Zhengbo Zhang1 , Chunluan Zhou2 , Zhigang Tu1 1Wuhan University 2Wormpex AI Research |
| Pseudocode | No | The paper describes its methods using prose and mathematical equations but does not include any explicitly labeled "Pseudocode" or "Algorithm" blocks. |
| Open Source Code | No | The paper does not include an explicit statement about releasing source code or a link to a code repository. |
| Open Datasets | Yes | We conduct comprehensive experiments on three popular benchmarks: Cityscapes [Cordts et al., 2016], Pascal VOC [Everingham et al., 2015], and ADE20K [Zhou et al., 2017]. |
| Dataset Splits | Yes | Cityscapes includes 5000 finely annotated images of driving scenes in cities. It consists of 2975, 500 and 1525 images for training, validation and testing, respectively. ... Pascal VOC composes of 1464 images for training, 1449 images for validation and 1456 images for testing. ... ADE20k is a challenging scene parsing dataset released by MIT, which contains 20K, 2K, 3K images with 150 classes for training, validation, and testing. |
| Hardware Specification | No | The numerical calculation was supported by the super-computing system in the Super-computing Center of Wuhan University. This statement is too general and does not specify any particular hardware components like GPU or CPU models. |
| Software Dependencies | No | We use the Pytorch platform to implement our method. This only mentions the platform without a specific version number or other software dependencies with versions. |
| Experiment Setup | Yes | Training Details. We use the Pytorch platform to implement our method. Following [Liu et al., 2019] , we train our student networks by mini-batch stochastic gradient descent (SGD) for 40000 iterations. We set the momentum and the weight decay as 0.9 and 0.0005, respectively. We apply the polynomial learning rate policy, and the learning rate is calculated as base lr 1 iter total iter power. The base learning rate and power are respectively set to 0.01 and 0.9. For the input images, we crop them to 512 512. The random scaling and random flipping are applied to augment the data. |