Learning from Future: A Novel Self-Training Framework for Semantic Segmentation

Authors: Ye Du, Yujun Shen, Haochen Wang, Jingjing Fei, Wei Li, Liwei Wu, Rui Zhao, Zehua Fu, Qingjie LIU

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

Reproducibility Variable Result LLM Response
Research Type Experimental We experimentally demonstrate the effectiveness and superiority of our approach under a wide range of settings.
Researcher Affiliation Collaboration Ye Du1,2 Yujun Shen3 Haochen Wang4 Jingjing Fei5 Wei Li5 Liwei Wu5 Rui Zhao5,6 Zehua Fu1,2 Qingjie Liu1,2 1 State Key Laboratory of Virtual Reality Technology and Systems, Beihang University 2 Hangzhou Innovation Institute, Beihang University 3 The Chinese University of Hong Kong 4 Institute of Automation, Chinese Academy of Sciences 5 Sense Time Research 6 Qing Yuan Research Institute, Shanghai Jiao Tong University, Shanghai, China
Pseudocode Yes We provide pseudo-codes to further illustrate how we implement Eq. (4) in Supplementary Material.
Open Source Code Yes Code is available at https://github.com/usr922/FST.
Open Datasets Yes In UDA segmentation, we use synthetic labeled images from GTAV [38] and SYNTHIA [39] as the source domain and use real images from Cityscapes [13] as the target domain. In addition, PASCAL VOC 2012 [15] is used for standard semi-supervised evaluation.
Dataset Splits Yes To simulate a semi-supervised setting, we randomly sample a portion (i.e., 1/4, 1/8, and 1/16) of images together with corresponding segmentation masks from the training set as the labeled data and treat the rest as the unlabeled samples.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory amounts, or cloud instance types) used for running its experiments.
Software Dependencies No The paper mentions optimizers (AdamW, SGD) and network architectures (Deep Lab V2, Deep Lab V3+, PSPNet, UPer Net) but does not provide specific version numbers for any software dependencies like Python, PyTorch, or TensorFlow.
Experiment Setup Yes In UDA segmentation, the model is trained with an Adam W [28] optimizer, a learning rate of 6 × 10−5 for the encoder and 6 × 10−4 for the decoder, a weight decay of 0.01, linear learning rate warmup with 1.5k iterations and linear decay afterwards. We train the model on a batch of two 512 × 512 random crops for a total of 40k iterations. The momentum u is set to 0.999. In semi-supervised segmentation, the model is trained with a SGD optimizer, a learning rate of 0.0001 for the encoder and 0.001 for the decoder, a weight decay of 0.0001. We train the model with 16 labeled and 16 unlabeled images per-batch for a total of 40 epochs.