How to retrain the network on ICDAR dataset ??

[AlwaysTrying]

Hi, I need some guide to run this project... some files are messing and the documentation is not enough... How can I reproduce the same results and get 85% as F-mesure??? The file demo.py does not exist, the data of ICDAR is not in data file... is there any tutoriel I can follow plz ??

[AlwaysTrying]

@MhLiao could you provide more details and explanations plz ??

[AlwaysTrying]

is it the script train_icdar13.py that i should run to train the data and generate the model ?? in this file is written:

The database file for training data. Created by data/text/create_data.sh

train_data = "path_to_train_lmdb/text_icdar_trainval_lmdb"

The database file for testing data. Created by data/text_10x/create_data.sh

test_data = "path_to_test_lmdb/text_icdar_test_lmdb"

but there is no text in /data !!! Am I on the wrong direction???

[MhLiao]

The data is too large. You can first generate ground truth in "xml" format, then use TextBoxes/data/VOC0712/create_data.sh to create the "lmdb" for training.

[liuyiyiyiyi]

when i run"python examples/TextBoxes/train_icdar13.py ",the error is occured: Traceback (most recent call last): File "examples/TextBoxes/train_icdar13.py", line 437, in net[reshape_name] = L.Reshape(net[conf_name], shape=dict(dim=[0, -1, num_classes])) File "/home/od/dod/SaicCorNetG_v2/python/caffe/net_spec.py", line 193, in getitem return self.getattr(item) File "/home/od/dod/SaicCorNetG_v2/python/caffe/net_spec.py", line 187, in getattr return self.tops[name] KeyError: 'mbox_conf'