murphypei / create-pascal-voc-dataset Goto Github PK

Refactor all the project ! Now it's more efficient and the structure is more clear.

中文版

1. Get your dataset annotations file, parse and save it to two files trainval_annotations.txt and test_annotations.txt, file name doesn't matter.

2. Your annotations file must have the format like this:

   image_full_path object1_class x1_min y1_min x1_max y1_max object2_class x2_min y2_min x2_max y2_max...

   • You can check examples/Train_annotation.txt file to understand the annotation format more clearly, this is the INRIA annotations file after my processing, your annotations file should be like this.

   • You should write your own dataset annotation process program, I just write for INRIA dataset and you can reference it in preprocess/inria_preprocess.py.

   • If I have more time, I will write more process program, you can send your requires in issues.

3. Edit your dataset config and run file；

   • Check the examples/inria_example.py to understand how to call the PASCALVOC07 class

   • Config your own information in your pascal voc dataset

   • Set the dataset directory, annotations file and output directory, then just run build, wait for your own pascal voc dataset.

• I have writen an example of the INRIA dataset:

python preprocess/inria_preprocess.py /path/to/INRIAPerson
python examples/inria_example.py /path/to/INRIAPerson /path/to/output

Anything can be send to issues and forgive my poor English...