Preprocessing4Semantic

Linux-cpp-python repository : Images Preprocessing for Semantic Segmentation

at least: opencv 2.4.3 or opencv 3.0.x

python : opencv-python, labelme(option) ...

Dataset Struct: example for KITTI

data_semantics ├── testing │ └── image_2

└── training ├── image_2

├── semantic

Folder named image_2 is Image sequence for RGB with the suffix ".png" ; Folder named semantic is Image sequence for annotation of semantic segmentation. And in this folder, each Image is uint8 8bit in single channel, with the suffix ".png", Each pixel value in image represents a semantic category ID. Attention : name of the same image in "image_2" folder or "semantic" folder is also the same.
Modification of documents :
- ./src/Preprocessing.cc: replace "strPatternRGB" and "strPatternSem" with your own folder's name.
- ./app/main.cpp: adjust percentage values for each category for image preprocessing.

mkdir NewRGB && mkdir NewSem
mkdir build && cd build && cmake .. && make -j4
../app/main <dataset path>

and new images in the NewRGB and NewSem folder.

At last , cp NewRGB -> image_2 ; NewSem->semantic and

python build_dataset.py --datasetPath <new-dataset-path> --outputPath <output-dataset>

prefix for new images and semantics:

ChenJiahao031008/Preprocessing4Semantic