repository·master·Indexed 19 days ago
https://github.com/tramac/fast-scnn-pytorchA PyTorch implementation of the Fast Semantic Segmentation Network (Fast-SCNN) designed for high-speed semantic segmentation tasks. It includes support for the Cityscapes dataset and provides scripts for training, evaluation, and visual demonstrations.
To set up the Fast-SCNN environment, ensure you have Python 3.x (Anaconda3 recommended) and PyTorch 1.0 installed. You can install PyTorch via conda using the following command:
conda install pytorch torchvision cudatoolkit=9.0 -c pytorchAfter installing dependencies, clone this repository and download the required datasets.
Fast-SCNN currently supports the Cityscapes dataset.
leftImg8bit_trainvaltest.zip (11GB) and gtFine_trainvaltest.zip (241MB) from the Cityscapes website../datasets/citys directory of the repository.To run evaluation on a previously trained model, execute the eval.py script.
python eval.pyTrain the model using the train.py script. You can specify parameters via command-line flags or by modifying the script directly. By default, the script expects the Cityscapes dataset in ./datasets/citys.
python train.py --model fast_scnn --dataset citysTo run a visual demonstration of the segmentation on a specific image, use the demo.py script with the --model and --input-pic flags.
python demo.py --model fast_scnn --input-pic './png/berlin_000000_000019_leftImg8bit.png'