Vladimir Zlobin

@xxlxx1 thank you for the reply.

Try checking a line of code where the error happens. Most probably this is https://github.com/Wovchena/text-detection-fots.pytorch/blob/04b969c87491630dce38cdb48932aa33115798c6/datasets.py#L153 That would mean that scandir() can't find your ch4_training_images. There is no much special about...

The pipeline crops images for training, so images just need to be not small. The net is inaccurate in detecting curved text.

Correct. The error on `high=(stretched.shape[0] // IN_OUT_RATIO) - OUT_SIDE` suggests that `stretched.shape[0]` can't be smaller than `OUT_SIDE`.

This happens because nothing is detected.

Looks correct

This boils down to whether `Polygon` from `shapely.geometry` can process different order of points. It defensively can't when the order is [0, 2, 1, 3] and it is handled by...

It only guarantees that annotations are correct from visualizer's point of view. But since `transform()` is a different function you cant say that it treats annotations in the exactly same...

Well, if you say ICDAR is in clockwise order, than clockwise order is correct order because the net works for ICDAR.

I think it doesn't meter: the points go through `minAreaRect()` which returns points in consistent order. `Polygon` also shouldn't care.