Vladimir Zlobin

I assume that happens in https://github.com/Wovchena/text-detection-fots.pytorch/blob/04b969c87491630dce38cdb48932aa33115798c6/datasets.py#L170 (that would be simpler if you left the full error with traceback) In that case you should build your tuple with `np.array()` (or `None`)...

Yes, I will be glad to review you patches

A) `matches = self.pattern.findall(line)[0]` parses a given `line` to extract coordinates of text and text label from a ground truth file. B) The error may occur if the `line` is...

Shape of `(0,)` means there is a one dimensional empty array. Such arrays shouldn't appear. You should figure out where that array comes from and handle it.

Training images are upscaled here: https://github.com/Wovchena/text-detection-fots.pytorch/blob/04b969c87491630dce38cdb48932aa33115798c6/datasets.py#L32 I think the model doesn't work for size of 384 because it was trained for a larger scale. You may also try visualizing raw...

It's generated by [`zip -jmq runs/u.zip res/*`](https://github.com/Wovchena/text-detection-fots.pytorch/tree/04b969c87491630dce38cdb48932aa33115798c6#test)

I don't know what causes the problem. It works for me with one image in the test folder. May be you have modified something before running. The strange thing is...

@Tianxiaomo there is no recognition branch. I am concerned only about the detection part.

> why is your weight is 417M? In the paper, the params is abut 20M. I don't know... I haven't had an idea to compare the value with the paper's...

Hi, @eyebies. `intersected_quad` is not supposed to be empty because of `for quad_id, quad in enumerate(quads)`. The loop iterates over the first dimension of `quads` array, so `quad` must have...