Dynamic Neural Representational Decoders for High-Resolution Semantic Segmentation
Authors: Bowen Zhang, Yifan liu, Zhi Tian, Chunhua Shen
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on the Cityscapes, ADE20K, and PASCAL Context datasets demonstrate the effectiveness and efficiency of our proposed method. 3 Experiments The proposed model is evaluated on three semantic segmentation benchmarks. The performance is measured in terms of intersection-over-union averaged across the present classes (m Io U). |
| Researcher Affiliation | Academia | Bowen Zhang, Yifan Liu, Zhi Tian, Chunhua Shen The University of Adelaide, Australia |
| Pseudocode | No | The paper describes the method using figures and text descriptions (e.g., Figure 1, Figure 3, Section 2.2) but does not include any formal pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not contain an explicit statement about releasing the source code for the described methodology or a direct link to a code repository for it. |
| Open Datasets | Yes | ADE20K [ZZP+17] is a dataset that contains more than 20K images exhaustively annotated with pixel-level annotation. ... PASCAL Context [MCL+14] is a dataset with 4, 998 images for training and 5, 105 images for validation. ... Cityscapes [COR+16] is a benchmark for semantic urban scene parsing. |
| Dataset Splits | Yes | ADE20K [ZZP+17] is a dataset that contains more than 20K images exhaustively annotated with pixel-level annotation. It has 20, 210 images for training and 2, 000 images for validation. ... Cityscapes [COR+16] is a benchmark for semantic urban scene parsing. The training, validation and test splits contain 2, 975, 500 and 1, 525 images with fine annotations, respectively. |
| Hardware Specification | Yes | The training and testing environment is on a workstation with four Volta 100 GPU cards. |
| Software Dependencies | No | The paper mentions using 'MMSegmentation' and Res Net backbones, but does not provide specific version numbers for software dependencies like PyTorch, TensorFlow, or CUDA. |
| Experiment Setup | Yes | The initial learning rate is set at 0.01, the weight decay is set to 0.0005 for Cityscapes and ADE20K. For PASCAL Context, the initial learning rate is 0.004 and the weight decay is 0.0001. We train ADE20K, PASCAL-Context and Cityscapes for 160k, 80k and 80k iterations, with the crop size of 512 512, 480 480 and 512 1024, respectively. ... For test time augmentation, we employ the horizontal flip and multi-scale inference. The scale factors are {0.5, 0.75, 1.0, 1.25, 1.5, 1.75}. |