microos / py-faster-rcnn-data-interface-generator

A text replacement based faster rcnn data interface generator

Python Faster-RCNN Data Interface Generator

The purpose of this tool is to generate dataset interface of Faster-RCNN(FRCN) that can help you to deploy training on your own dataset as fast as possible.

Usage

This tool will generate dataset interface like pascal_voc.py, voc_eval.py and factory.py. Also a set of net prototxt files like py-faster-rcnn/models/pascal_voc. And a bash script similar to py-faster-rcnn/experiments/scripts/faster_rcnn_end2end.sh

Before using this generator, please make sure you have already set up a voc-like tree structure of your dataset(in your py-faster-rcnn/data/ folder):

YOUR_DATASET_devkit
    ├── data
    ├── Annotations
    │   └── *.xml
    │
    ├── ImageSets
    │   └── Main
    │       ├── train.txt
    │       └── test.txt
    └── JPEGImages
        └── *.jpg

Then you can run the generate.py to generate interface, prototxt and script files

Arguments:

--froot: absolute path to your faster-rcnn root.
--idname: name your dataset, whatever you want.s It will be used in the future training.
--cls: class names in your dataset, use commas to split them (e.g: cat,dog,tiger)
--dvkt: absolute path to your dataset devkit

Final Step

Please note that the tool will do everything automatically for you except overwriting the factory.py file.

In case of changing the factory.py file without any backup, the tool will not overwrite it but create a new <idname>_factory.py file for you.

You have 2 options to complete the last step manually:

1. Backup the original factory.py file and replace it with the generated one.
2. Copy the code block displayed in the task summary into the factory.py.

Reference

deboc's tutorial