R²CNN: Rotational Region CNN for Orientation Robust Scene Text Detection

Abstract

This is a caffe re-implementation of R²CNN: Rotational Region CNN for Orientation Robust Scene Text Detection.

This project is modified from py-R-FCN, and inclined nms and generate rotated box component is imported from EAST project. Thanks for the author's(@zxytim @argman) help. Please cite this paper if you find this useful.

Contents

1. Abstract
2. Structor
3. Installation
4. Demo
5. Test
6. Train
7. Experiments
8. Furthermore

Structor

Code structor

.
├── docker-compose.yml
├── docker // docker deps file
├── Dockerfile // docker build file
├── model // model directory
│   ├── caffemodel // trained caffe model
│   ├── icdar15_gt // ICDAR2015 groundtruth
│   ├── prototxt // caffe prototxt file
│   └── imagenet_models // pretrained on imagenet
├── nvidia-docker-compose.yml
├── logs
│   ├── submit // original submit file
│   ├── submit_zip // zip submit file
│   ├── snapshots
│   └── train
│       ├── VGG16.txt.*
│       └── snapshots
├── README.md
├── requirements.txt // python package
├── src
│   ├── cfgs // train config yml
│   ├── data // cache file
│   ├── lib
│   ├── _init_path.py
│   ├── demo.py
│   ├── eval_icdar15.py // eval 2015 icdar dataset F-meaure
│   ├── test_net.py
│   └── train_net.py
├── demo.sh
├── train.sh
├── images // test images
│   ├── img_1.jpg
│   ├── img_2.jpg
│   ├── img_3.jpg
│   ├── img_4.jpg
│   └── img_5.jpg
└── test.sh // test script

Data structor

It should have this basic structure

ICDARdevkit_Root
.
├── ICDAR2013
├── merge_train.txt  // images list contains ICDAR2013+ICDAR2015 train dataset, then raw data augmentation the same as the paper
├── ICDAR2015
│   ├── augmentation // contains all augmented images
│   └── ImageSets/Main/test.txt // ICDAR2015 test images list

Installation

Install caffe

It is highly recommended to use docker to build environment. More about how to configure docker, see Running with Docker If you are familiar with docker, please run

    1. nvidia-docker-compose run --rm --service-ports rrcnn bash
    2. bash ./demo.sh

If you don't familiar with docker, please follow py-R-FCN to install caffe.

Build

    cd src/lib && make
    

Download Model

1. please download VGG16 pre-trained model on Imagenet, place it to model/imagenet_models/VGG16.v2.caffemodel.
2. please download VGG16 trained model by this project, place it model/caffemodel/TextBoxes-v2_iter_12w.caffemodel.

Demo

It is recommended to use UNIX socket to support GUI for docker, plesase open another terminal and type:

    xhost + # may be you need it when open a new terminal
    # docker-compose.yml: mount host  volume : /tmp/.X11-unix to docker volume: /tmp/.X11-unix  
    # pass DISPLAY variable to docker container so host X server can display image in docker
    docker exec -it -e DISPLAY=$DISPLAY ${CURRENT_CONTAINER_ID} bash
    bash ./demo.sh

Test

Single Test

    bash ./test.sh

Multi-scale Test

    # please uncomment two lines in src/cfgs/faster_rcnn_end2end.yml
    SCALES: [720, 1200]
    MULTI_SCALES_NOC: True
    # modify src/lib/datasets/icdar.py to find ICDAR2015 test data, please refer to commit @bbac1cf
    # then run
    bash ./test.sh

Train

Train data

• Mine: ICDAR2013+ICDAR2015 train dataset, and raw data augmentation, at last got 15977 images.
• Paper: ICDAR2015 + 2000 focused scene text images they collected.

Train commands

1. Go to ./src/lib/datasets/icdar.py, modify images path to let train.py find merge_train.txt images list.
2. Remove cache in src/data/*.pkl or you can load cached roidb data of this project, and place it to src/data/

    # Train for RRCNN4-TextBoxes-v2-OHEM
    bash ./train.sh

note: If you use USE_FLIPPED=True&USE_FLIPPED_QUAD=True, you will get almost 31200 roidb.

Experiments

Mine VS Paper

Appendixes

Furthermore

You can try Resnet-50, Resnet-101 and so on.