Class Guided Channel Weighting Network for Fine-Grained Semantic Segmentation

Authors: Xiang Zhang, Wanqing Zhao, Hangzai Luo, Jinye Peng, Jianping Fan3344-3352

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on PASCAL VOC 2012 and six fine-grained image sets show that our proposed CGCWNet has achieved state-of-the-art results. and Experimental Results and Analysis
Researcher Affiliation Academia Northwest University, Xi an, China {Zhang Xiang2015@stumail., zhaowq@, hzluo@, pjy@, jfan@}nwu.edu.cn
Pseudocode No The paper contains architectural diagrams (flowcharts) but no formal pseudocode or algorithm blocks.
Open Source Code No The paper does not provide an explicit statement or link for open-source code availability.
Open Datasets Yes We extend the fine-grained image classification datasets (i.e., FGVC Aircraft (Maji et al. 2013), CUB-2002011 (Xiao et al. 2015), Stanford Cars (Krause et al. 2013), and Orchid Plant) to fine-grained segmentation datasets. In our experiments, the proposed CGCWNet has achieved state-of-the-art results on PASCAL VOC 2012 (Hariharan et al. 2011) and expanded six finegrained image sets. and The PASCAL VOC 2012 is a semantic segmentation benchmark with 20 foreground object classes and one background class. The dataset is augmented by the extra labellings provided by (Hariharan et al. 2011)
Dataset Splits Yes The dataset is augmented by the extra labellings provided by (Hariharan et al. 2011), which has 10,582, 1,449, and 1,456 images for network training, validation, and testing, respectively. and Table 1: Statistics of fine-grained datasets used in this paper.
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory) were mentioned for running experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions). It mentions frameworks/models used but not their specific versions.
Experiment Setup Yes For network training, we use min-batch stochastic gradient descent (SGD) optimizer with the batch size 6, initial learning rate 4e 3, weight decay 0.0002, and momentum 0.9 for Stanford Cars, CUB-200-2011, FGVC Aircraft, Orchid Plant, and PASCAL VOC 2012 image sets. Following some previous works (Chen et al. 2018a; Yu et al. 2018b), we use the poly learning rate policy where the learning rate is multiplied by the factor (1 iter/max iter)0.9. In the DEGF module, the values of r and ε are first determined by grid search on the validation set, and then we use the same parameters to train the CGCWNet. In our network, C and b C are set to 2048 and 512 respectively. The loss weights λa, λc, and λf in Eq. (5) are set to 0.4, 0.6, and 1.0 respectively.