Learning Generalized Segmentation for Foggy-Scenes by Bi-directional Wavelet Guidance
Authors: Qi Bi, Shaodi You, Theo Gevers
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Large-scale experiments are conducted on four foggy-scene segmentation datasets under a variety of interesting settings. The proposed method significantly outperforms existing directly-supervised, curriculum domain adaptation and domain generalization segmentation methods. |
| Researcher Affiliation | Academia | Computer Vision Research Group, University of Amsterdam, Netherlands {q.bi, s.you, th.gevers}@uva.nl |
| Pseudocode | No | The paper includes figures, equations, and descriptions of the method, but no clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Source code is available at https://github.com/Bi Qi WHU/BWG. |
| Open Datasets | Yes | City Scapes (Cordts et al. 2016) is a commonly-used semantic segmentation datasets for driving-scenes. (...) Foggy-City Scapes (Sakaridis, Dai, and Van Gool 2018) contains 550 synthetic foggy images in total (...). Foggy Zurich (Sakaridis et al. 2018) contains 3,808 realworld foggy road scenes from the Zurich city. (...) Foggy Driving (Sakaridis, Dai, and Van Gool 2018) has 101 real-world foggy road-scenes images. (...) Adverse Conditions Dataset with Correspondences (ACDC) (Sakaridis, Dai, and Van Gool 2021) has 4006 driving-scene segmentation samples under adverse conditions. |
| Dataset Splits | Yes | City Scapes (Cordts et al. 2016) is a commonly-used semantic segmentation datasets for driving-scenes. It has 2965 training samples and 500 validation samples, with 19 common scene categories in driving-scenes. (...) Adverse Conditions Dataset with Correspondences (ACDC) (Sakaridis, Dai, and Van Gool 2021) has 4006 driving-scene segmentation samples under adverse conditions. 1000 of them are foggy images. Data split for training, validation and testing is 4:1:5. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU models, CPU types, or memory specifications used for running the experiments. It only mentions the use of a Swin-base Transformer backbone. |
| Software Dependencies | No | The paper mentions that "the Adam optimizer is used" and that "All the loss and hyper-parameter settings keep the same as the original Mask2Former (Cheng et al. 2022)", but it does not specify versions for programming languages (e.g., Python), deep learning frameworks (e.g., PyTorch, TensorFlow), or other software libraries with version numbers. |
| Experiment Setup | Yes | All the loss and hyper-parameter settings keep the same as the original Mask2Former (Cheng et al. 2022) without any additional fine-tuning. By default, the Adam optimizer is used with an initial learning rate of 1 10 4. The weight decay is set 0.05. The training terminates after 50 epochs. |