Segment or detect objects on images using tensorflow. Models are from different sources
launcher.py [-h] -t MODEL_TYPE [-vf VIDEO_FILE] [-di IN_DIRECTORY] -do OUT_DIRECTORY
This tool supports video using the -vf
flag. Input videos can be in any format supported by OpenCV. The output video is written using the MJPEG codec.
python3 launcher.py -di /dir/with/images -do /output/directory -t segmentationcs
To use this model, you need to download one pretrained model from [https://github.com/tensorflow/models/blob/master/research/deeplab/g3doc/model_zoo.md].
Tested only with deeplabv3_mnv2_cityscapes_train_2018_02_05.tar.gz
python3 launcher.py -di /dir/with/images -do /output/directory -t segmentation
python3 launcher.py -di /dir/with/images -do /output/directory -t detection