argman

I think small objects maybe filtered by nms during generating proposals, you may check the nms parameter

thanks for your reply, check must be added to make sure a char is supported by this font ``` text_size = font.get_metrics(text) supported = len(filter(lambda t: t is None, text_size))...

there is a script in icdar's website

you need first register , when you submit result , you can find it

.............can you provide some sample images?

currently EAST is more suitable for detecting words(or short text), if you want to detect long texts, you can resize the width side. If your images contain all like this,...

no, because the receptive field of CNN is too small

the c++ nms is in python3, you can use then python version

I think use a downsample rate of 1/4 is to save computation

you can load the weights from the ckpt i provide, just replace the imagenet cktp with icdar ckpt