Self-Decoupling and Ensemble Distillation for Efficient Segmentation

Authors: Yuang Liu, Wei Zhang, Jun Wang

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The extensive experiments on three public segmentation datasets demonstrate the superiority of our approach and the efficacy of each component in the framework through the ablation study.
Researcher Affiliation Academia Yuang Liu, Wei Zhang*, Jun Wang* School of Computer Science and Technology, East China Normal University {frankliu624, zhangwei.thu2011, wongjun}@gmail.com
Pseudocode No No structured pseudocode or algorithm blocks were found. The methods are described in narrative text and mathematical equations.
Open Source Code No The paper does not provide a link or an explicit statement regarding the availability of its source code.
Open Datasets Yes Pascal VOC 2012 (Everingham and Winn 2011) is a visual object segmentation dataset that that consists of 21 classes... Cityscapes (Cordts et al. 2016) contains 5000 highresolution images... Cam Vid (Brostow et al. 2008) is an automotive dataset...
Dataset Splits Yes Pascal VOC 2012 ... 10582/1449/1456 images for train/val/test. Cityscapes ... 2975 fine annotation images for training, 500 for validation, and 1525 for testing. Cam Vid ... 367/101/233 images for train/val/test
Hardware Specification Yes Our approach is implemented by PyTorch with two NVIDIA 2080Ti GPUs.
Software Dependencies No The paper mentions 'PyTorch' but does not provide specific version numbers for it or any other software dependencies.
Experiment Setup Yes The student networks are optimized by mini-batch stochastic gradient descent (SGD) with the momentum (0.9) and the weight decay (0.0001). We set the initial learning rate as 0.01 and use poly learning rate decay where the initial learning rate is multiplied by (1 iter max iters)0.9 after each iteration. The number of the total training iterations is 30K/45K/5K for VOC/Cityscapes/Cam Vid with a batch size of 8. The hyperparameter τ, α and β are set to 10, 0.5 and 100 by default, respectively.