Practical overview
When it fits
Use it when you need a codebase for a Cityscapes training workflow, model evaluation, or an image segmentation demo, and can work with Python and PyTorch.
Who it is for
- Developers exploring or adapting a PyTorch semantic-segmentation implementation.
- Researchers or practitioners training and evaluating Fast-SCNN on Cityscapes.
Documented uses
- Train Fast-SCNN using the documented Cityscapes dataset workflow.
- Evaluate a trained network with the repository's evaluation script.
- Generate a segmentation demo from an input image.
Requirements & limits
The repository is public and identifies its license as Apache-2.0.
- The documented training dataset is Cityscapes; VOC and ADE20K are described as future work.
- The README specifies Python 3.x and PyTorch 1.0 and gives an example installation command using torchvision and CUDA Toolkit 9.0.
- The repository separates training, evaluation, and demo tasks into scripts.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Train Fast-SCNN with Cityscapes
Prepare the documented environment and dataset, then start training with the repository's example command.
Before you start
- Python 3.x; Anaconda3 is recommended.
- PyTorch 1.0; the example installation command also uses torchvision and CUDA Toolkit 9.0.
- The cloned repository and Cityscapes files: leftImg8bit_trainvaltest.zip and gtFine_trainvaltest.
- Place the downloaded Cityscapes dataset in ./datasets/citys.
Steps
Install Python 3.x and PyTorch 1.0, using the PyTorch website to select an environment-specific installation command.
Clone the repository and download the two specified Cityscapes archives.
Place the dataset in ./datasets/citys.
Run the command below; adjust parameters through flags or in train.py as needed.
python train.py --model fast_scnn --dataset citys
Expected outcome
The repository's training script is invoked for Fast-SCNN with the Cityscapes dataset.
Guide limitations
- The README documents Cityscapes as the currently supported training dataset; VOC and ADE20K are listed as future work.
- The instructions describe the training command but do not establish that it has been run successfully in a particular environment.
Run the documented image demo
Invoke the demo script on the repository's example input image.
Before you start
- Python 3.x and PyTorch 1.0, as specified in the repository setup.
- The cloned repository and its example image at ./png/berlin_000000_000019_leftImg8bit.png.
Steps
Set up Python and PyTorch according to the repository installation instructions.
From the repository, run the command below.
python demo.py --model fast_scnn --input-pic './png/berlin_000000_000019_leftImg8bit.png'
Expected outcome
The documented demo command requests a Fast-SCNN segmentation for the example input image.
Guide limitations
- The README provides an example command but does not specify additional input formats or deployment behavior.
- The command is documented guidance, not a verified execution result.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 17 October 2026
- Original website title
- Tramac/Fast-SCNN-pytorch · Official homepage title · retained in its source language