Can you release lable of multi-lingual scene text dataset without split? - Githubissues

eragonruan / text-detection-ctpn

text detection mainly based on ctpn model in tensorflow, id card detect, connectionist text proposal network

MIT License

3.43k stars 1.33k forks source link

Can you release lable of multi-lingual scene text dataset without split? #291

Open hcnhatnam opened 5 years ago

hcnhatnam commented 5 years ago

I want to traning with fixed 8-pixel width instead of 16-pixel but lable of dataset was splited 16-pixel.

hcnhatnam commented 5 years ago

@eragonruan @banjin-xjy help me,pls.

eragonruan commented 5 years ago

@hcnhatnam hi, check this link for the original training data we used. or you can download from icdar website

hcnhatnam commented 5 years ago

@eragonruan Sorry but I'm foreigner.I am almost too hard to use baidu and i think you just release lable folder in Google Drive. Thank you.

eragonruan commented 5 years ago

https://drive.google.com/file/d/1gIq2E2PRJnMGMmGskHS-m3-LVPt5tor7/view?usp=sharing

hcnhatnam commented 5 years ago

Thank you so much @eragonruan.

hcnhatnam commented 5 years ago

@eragonruan Sorry but I think there are a few mistakes here. I check groundtruth of lable that you shared but the coordinates in the file are not correct. Example: gt_109_icdar13.txt of lable folder:

1137.0,1264.0,2093.0,1264.0,2093.0,1446.0,1137.0,1446.0,"syncmaster"
2186.0,1284.0,2558.0,1284.0,2558.0,1431.0,2186.0,1431.0,"213t"

but 109_icdar13.png in mlt dataset has size 912x608 Tag "syncmaster", "213t" are correct.

NamNguyenThanh commented 5 years ago

@eragonruan Please check it for us. I very appreciate that!