RaiuSuinTawo / DFS

Dual-input Fusion Segmenter: an effective framework for semantic segmentation of detritus from river sands

Dual-input Fusion Segmenter: an effective framework for semantic segmentation of detritus from river sands

Models Available for Training

Our repository includes the DFS model and various semantic segmentation models used for comparative experiments, such as fcn, unet, ccnet, deeplab, segformer, and more (refer to geologyseg/model and geology_seg/model/deeplabv3plus). Additionally, we explore different backbones for semantic segmentation, including resnet, hrnet, xception, etc. (found in geology_seg/model/deeplabv3plus/backbone).

For a detailed view of the code for our DFS model, please see geology_seg/model/deeplabv3plus/new_arch.py and geology_seg/model/deeplabv3plus/new_utils.py. The code for model training is available in geology_seg/train.py, while the code for model prediction can be found in geology_seg/predict.py.

Experimental Data

The experimental data used in the paper is located in .txt files with val and pred prefixes under geology_seg and val_texts, along with a .csv file under geology_seg/results_coarse.

Remaining

Kindly utilize the pre-existing scripts in the repository to retrain the model and execute the required tasks.