search

Belval / CRNN

A TensorFlow implementation of https://github.com/bgshih/crnn
MIT License
297 stars 101 forks source link

CHAR_VECTOR is correct? #60

Open yoyoyo-yo opened 4 years ago

yoyoyo-yo commented 4 years ago

Thank you for your great job!

I tried your pretrained model, but got error.

Traceback (most recent call last): File "/Users/yoshitonagaoka/dev/miniconda3/envs/deepscholar-crnn/lib/python3.7/site-packages/tensorflow/python/client/session.py", line 1356, in _do_call return fn(*args) File "/Users/yoshitonagaoka/dev/miniconda3/envs/deepscholar-crnn/lib/python3.7/site-packages/tensorflow/python/client/session.py", line 1341, in _run_fn options, feed_dict, fetch_list, target_list, run_metadata) File "/Users/yoshitonagaoka/dev/miniconda3/envs/deepscholar-crnn/lib/python3.7/site-packages/tensorflow/python/client/session.py", line 1429, in _call_tf_sessionrun run_metadata) tensorflow.python.framework.errors_impl.InvalidArgumentError: Assign requires shapes of both tensors to match. lhs shape= [512,71] rhs shape= [512,70] [[{{node save/Assign_1}}]]

I fixed CHAR_VECTOR in run.py like below, and worked well.

CHAR_VECTOR = "0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ-'.!?,\"&"

CHAR_VECTOR = "0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ-'.!?,&"

Please tell the correct CHAR_VECTOR for your pretrained model.

mrtpk commented 4 years ago

Please see the issue #61