DeeplavV3 image segmentation with UE5 generated dataset and Tensorflow.js
https://www.unrealengine.com/marketplace/en-US/product/city-sample
I use UE5 tech demo "city sample" asset for generate train, validation dataset
and this repo also contains dataset at model/unreal-car-dataset
| image | mask |
|---|---|
![]() |
![]() |
The data set looks like this. In fact, since the class value of the mask has a gray scale of 1, the actual data cannot be distinguished by the eye unless the gamma is raised.
Provides a web app using tensorflow js.
Learned models are already included and can be used as soon as you run them.
model: contains model train, model save, model convert, train/validation dataset
- model/train_model.py - train model and generate h5
- model/use_model.py - test model with validation data
- model/tfjs_modelgen.py - convert model compatible to tensorflow.js
webapp: react web app that load the model using tenserflow.js
- webapp/public/tfjs-model - converted model for tenserflow.js


