search

MichalBusta / DeepTextSpotter

285 stars 101 forks source link

retrain failed #82

Open NandaPRomadhona opened 5 years ago

NandaPRomadhona commented 5 years ago

Hi, Thank you for sharing this code.

We got an error when retrain and run train.py with my own data.

After tracing code train.py we found that program failed to execute line code optim2.step(1) in method process_batch . We were following solution on issue #53 to separate the train code to be ctc_train.py that generate OCR and det_train.py that generate RPN. And the error was same. We got an error when execute line code optim2.step(1) in det_train.py . The error message was, "floating point exceptionally exception" something like that. We tried to change optim2.step(1) to be optim2.step(0) and the line code passed but still error on line code o_image = net.layers[0].get_image_file_name(bid) with error message, "segmentation fault (core dumped)" something like that.

We have no idea how to fix it? and What does '1' mean on line code optim2.step(1) ?

MichalBusta commented 5 years ago

Hi,

We got an error when retrain and run train.py with my own data.

After tracing code train.py we found that program failed to execute line code optim2.step(1) in method process_batch . We were following solution on issue #53 to separate the train code to be ctc_train.py that generate OCR and det_train.py that generate RPN. And the error was same. We got an error when execute line code optim2.step(1) in det_train.py . The error message was, "floating point exceptionally exception" something like that. We tried to change optim2.step(1) to be optim2.step(0) and the line code passed but still error on line code o_image = net.layers[0].get_image_file_name(bid) with error message, "segmentation fault (core dumped)" something like that.

it looks like data feeding problem, check tiny.prototxt:

layer { name: "data" type: "OnDiskData" top: "data" input_param { shape: { dim: 1 dim: 1 dim: 32 dim: 32 } } data_param { source: "/home/busta/data/SynthText/train_icdar.txt" } top: "gt_boxes" top: "gt_labels" top: "line_boxes" top: "line_labels" } line: data_param { source: "/home/busta/data/SynthText/train_icdar.txt" } there is unchecked input - if file does not exists, you will get segmentation fault.

We have no idea how to fix it? and What does '1' mean on line code optim2.step(1) ?

1 = how many steps should optimizer take.

NandaPRomadhona commented 5 years ago

Hi, thanks. Those issues has been fixed. I was successfully run train.py and produced caffemodel. But the type model is different, your model has type "ARC archive" and my model has type "PCX image". So my model could not be executed on demo.py Do you have any suggestion?

MichalBusta commented 5 years ago

Uff - no idea :(