Transferable Semi-Supervised Semantic Segmentation
Authors: Huaxin Xiao, Yunchao Wei, Yu Liu, Maojun Zhang, Jiashi Feng
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct experiments on the PASCAL VOC 2012 dataset, and in case of only 50% (30%) categories with pixel-level annotations, our proposed model achieves 96.5% (91.4%) performance of the fully-supervised baseline. |
| Researcher Affiliation | Academia | 1Department of System Engineering, National University of Defense Technology 2Department of ECE, National University of Singapore 3Beckman Institute, University of Illinois at Urbana-Champaign |
| Pseudocode | No | No explicit pseudocode or algorithm blocks were found. |
| Open Source Code | No | The paper does not contain any explicit statement about providing open-source code or a link to a code repository. |
| Open Datasets | Yes | We evaluate the performance of the proposed model on the PASCAL VOC 2012 benchmark (Everingham et al. 2014) which contains one background category and 20 object categories. The training set contains 10,582 images with pixel-level annotations, which is extended by Hariharan et al. (2011). |
| Dataset Splits | Yes | We evaluate the performance in terms of mean Intersection over Union (m Io U) on other two subsets, i.e., validation and test, including 1,449 and 1,456 images respectively. |
| Hardware Specification | Yes | All the experiments are performed on NVIDIA TITAN X PASCAL GPU with 12G memory. |
| Software Dependencies | No | The paper mentions using 'Deep Lab Large FOV' and 'VGG-16 model' but does not specify software dependencies like Python, PyTorch/TensorFlow, or CUDA versions with version numbers. |
| Experiment Setup | Yes | For the training of L-Net, we convert the semantic label maps from strong categories to a binary mask. We take a mini-batch size of 30, in which patches of 321 321 pixels are randomly cropped from images. We totally perform 30 epochs for training the L-Net with an initial learning rate of 5e-8. Momentum and weight decay are set to 0.9 and 0.0005 respectively. |