The difference between SynText and ICDAR13

[lucas946]

I see the format of SynText is different from ICDAR13, the coordinate of boxes in SynText is about 4 points (e.g. [(x1, x2, x3, x4),(y1,y2,y3,y4)]), but the coordinate of boxes in ICDAR13 is about 2 points (x1, y1, x2, y2). I think the difference will cause the change of network structure (e.g. the kernel nums or loss functions and so on). I don't understand how to pre-train and train model in two different network structure, can you tell me how to solve this problem?

[MhLiao]

All the formats are converted to (xmin, ymin, xmax, ymax), which is the minimum rectangle that covers the quadrilateral.

[lucas946]

Your answer help me a lot, Thanks~