Skip to content
Branch: master
Find file History
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
..
Failed to load latest commit information.
examples
lib
Dockerfile
README.md
docker.requirements.txt
run.sh
setup.sh

README.md

src

This repository contains scripts to run necessary transforms and post-processing for the data produced from UE4 in preparation for training ML classifiers.

setup

From a bash or zsh shell (no fish support) inside this folder ('src'), run:

source setup.sh

This command will create a Docker image with the necessary dependencies installed, and spin up a bash shell inside a Docker container created from that image. The Docker image will not rebuild if it already exists in your local image list.

UE4 to Supervisely Workflow

Annotations are exported from UE4 simply as an image that consists of another render pass, where pixels that include an object of interested are a non-black colour. Each separate object is coded with a different colour. There is currently no semantic to encode multiple different objects of interest in an annotation that comes from UE4.

Ensure that the exported images are organised with the following folder structure:

. +-- img
| +-- img1.png
| +-- img2.png
+-- raw_ann
| +-- img1.png
| +-- img2.png

The original image should be in the 'img' folder, and the mask should be a file of the same name in the 'raw_ann' folder.

Once you have confirmed that the folder is organised correctly, adjust the variables that are templated in run.sh for input and output paths appropriately. Then, inside the docker container (entered via the command above), run:

sh run.sh

This will convert the masks that UE4 produces to Supervisely's annotation format. See the variables that are templated inside run.sh and adjust accordingly for different input and output paths, dataset names, and label names.

You can’t perform that action at this time.