Akimoto-Cris / keras-frcnn

keras-frcnn

Keras implementation of Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks. cloned from https://github.com/yhenon/keras-frcnn/

USAGE:

• Both theano and tensorflow backends are supported. However compile times are very high in theano, and tensorflow is highly recommended.

• main.py can be used to train and test a model. To train on Pascal VOC data, simply do: python main.py --mode train -p /path/to/pascalvoc/.

• the Pascal VOC data set (images and annotations for bounding boxes around the classified objects) can be obtained from: http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar

• simple_parser provides an alternative way to input data, using a text file. Simply provide a text file, with each line containing:

  filepath,x1,y1,x2,y2,class_name

  For example:

  /data/imgs/img_001.jpg,837,346,981,456,cow

  /data/imgs/img_002.jpg,215,312,279,391,cat

  The classes will be inferred from the file. To use the simple parser instead of the default pascal voc style parser, use the command line option -o simple. For example python main.py --mode train -o simple -p annotation.txt.

• Training will write weights to disk to an hdf5 file, as well as all the setting of the training run to a pickle file. These settings can then be loaded in testing part. Thus in testing, only these 4 arguments are required:

  • --mode, --config_filename, --path, --num_rois

• Test mode in main.py can be used to perform inference, given pretrained weights and a config file. Specify a path to the folder containing images: python test_frcnn.py -p /path/to/test_data/

• Data augmentation can be applied by specifying --hf for horizontal flips, --vf for vertical flips and --rot for 90 degree rotations

• In training, the value of output_weight_path must contains a pair of empty brace, you could refer to the default value: weights\\model_frcnn_{}.

• In testing, the value of input_weight_path is NOT the file name, it should be the prefix of the weights files output by training process, without classifier/rpn.h5 suffix. for example, if the weight files are weights\model_frcnn_resnet50classifier.h5 and weights\model_frcnn_resnet50rpn.h5, then the value of input_weight_path should be weights\model_frcnn_resnet50.

• The value of --path in training should be the path to annotation file, e.g. annotation.txt, while in testing the value should be a directory that includes the images to be tested.

SIMPLE EXAMPLE:

• Train: python main.py --mode train -o simple -p annotation.txt -hf -rot -c config_own.pickle
• Testing: python main.py --mode train -o simple -p Image\\001

NOTES:

• config.py contains all settings for the train or test run. The default settings match those in the original Faster-RCNN paper. The anchor box sizes are [128, 256, 512] and the ratios are [1:1, 1:2, 2:1].
• The theano backend by default uses a 7x7 pooling region, instead of 14x14 as in the frcnn paper. This cuts down compiling time slightly.
• The tensorflow backend performs a resize on the pooling region, instead of max pooling. This is much more efficient and has little impact on results.