CRAFT-pytorch
CRAFT-pytorch copied to clipboard
input image size
What is the maximal image input size for craft text detection?
Is the success rate for text detection dependent on the image size?
@SkygirlLuna , Inference time of the model increase when you increase the input image size. Also, this other parameters like link_threshold, text_threshold, low_text plays a role in accurate text detection. I used below settings to get optimal speed and accuracy.
size = max(image.shape[0],image.shape[1],640)
# Reshaping to the nearest size
size = min(size,2560)
text_threshold=0.5,
link_threshold=0.1,
low_text=0.30