Mayank Kumar Singh
Mayank Kumar Singh
Also the code I gave above (https://github.com/clovaai/CRAFT-pytorch/issues/3#issuecomment-505903264) breaks down if the character quadrilateral is not valid. To check you can use - from shapely.geometry import Polygon Polygon(bbox.reshape([4, 2]).astype(np.int32)).is_valid: If this...
> @mayank-git-hub > How appropriate is the value of spread? > I think 1 is the most similar to the picture in the paper. > what do you think..? >...
You are assuming that the character bbox would always be horizontal which is not the case. Also, if you create the Gaussian heatmap everytime for a new character, it would...
@hanish3464 You could try this explanation of the perspective transform. https://www.pyimagesearch.com/2014/08/25/4-point-opencv-getperspective-transform-example/ They are trying to bring a skewed image back to horizontal rectangle using four_point_transform, while I am doing the...
Quadrilateral instead of rectangle, not polygon
> @mayank-git-hub hmm the follow you export heatmap is bellow ? > Step1: read annotation get position for each char > Step2: crop image with the point from step1 and...
I think the Gaussian heat map you are using has a very small standard deviation because of which the characters are not getting connected by the affinity. Try increasing the...
Another approach which works is to first train the model on a clean dataset and once the model is trained, freeze the model parameters and add two enhancement blocks to...