The code of Kernel Proposal Network for Arbitrary Shape Scene Text Detection for review only.
PyTorch >= 1.2.0 torchvision
cycler easydict matplotlib numpy opencv-python Pillow Polygon3 scipy Shapely tensorboardX tqdm
Put your texting images in the floders of
"../../data_model/data/total-text-mat/Images/Test/"
"../../data_model/data/ctw1500/test/text_image/"
"../../data_model/data/Icdar2015/Test/"
You can change the relative path "../../data_model/" in util/config.py: config.data_model_path = "../../data_model/"
You can get the model weight at model. And unzip them in current path.
You can run the code by:
sh eval_totaltext.sh [gpu ID]
sh eval_ctw1500.sh [gpu ID]
sh eval_IC15.sh [gpu ID]
if you want to view the separate text instances, you can add "--eval_vis True" in the above ".sh", it will show the separate text instances with the OpenCV "cv2.imshow".
The results will be stored in the floder "vis".