johnsmithm / handwritten-tf-1.0 Goto Github PK

hello, @johnsmithm , I found you use slices to cut the original input image to a pack of slices, why not just use CNN stride param to represent the sliding windows?

And when you use slices, how to handle the labels？ It's hard to align each slice with a correct label.

Why is the loss so high in the training? I have resized images to (36,90). it's around 76-88.

what's wrong?
And training exit with only 20 steps??

Hi I created tfrecords. Then i ran tried to train the model. But I am getting below error. I tried to debug the code but i am not able to do so.

Error:
INFO:tensorflow:Error reported to Coordinator: <class 'tensorflow.python.framework.errors_impl.InvalidArgumentError'>, Name: , Key: imageInput, Index: 0. Number of float values != expected. Values size: 3456000 but output shape: [3240]
[[Node: test_input/ParseExample_1/ParseExample = ParseExample[Ndense=2, Nsparse=1, Tdense=[DT_FLOAT, DT_INT64], dense_shapes=[[3240], [1]], sparse_types=[DT_INT64], _device="/job:localhost/replica:0/task:0/cpu:0"](test_input/ReaderReadUpToV2_1:1, test_input/ParseExample_1/ParseExample/names, test_input/ParseExample_1/ParseExample/sparse_keys_0, test_input/ParseExample_1/ParseExample/dense_keys_0, test_input/ParseExample_1/ParseExample/dense_keys_1, test_input/ParseExample_1/Const, test_input/ParseExample_1/Const_1)]]

Caused by op u'test_input/ParseExample_1/ParseExample', defined at:
File "train.py", line 818, in
app.run()
File "/home/nikhil/anaconda/envs/tfone/lib/python2.7/site-packages/tensorflow/python/platform/app.py", line 48, in run
_sys.exit(main(_sys.argv[:1] + flags_passthrough))
File "train.py", line 806, in main
FLAGS.export_model_steps).run(start_new_model=FLAGS.start_new_model)
File "train.py", line 435, in run
saver = self.build_model(self.model, self.reader)
File "train.py", line 699, in build_model
num_epochs=FLAGS.num_epochs)
File "train.py", line 280, in build_graph
num_epochs=num_epochs,nameT='test'))
File "train.py", line 200, in get_input_data_tensors
reader.prepare_reader(filename_queue,batch_size) for _ in range(num_readers)
File "/media/nikhil/Data/Experiments/Handwritten text Extraction/handwriting Extraction Codes/handwritten-tf-1.0-master/readers.py", line 81, in prepare_reader
return self.prepare_serialized_examples(serialized_examples)
File "/media/nikhil/Data/Experiments/Handwritten text Extraction/handwriting Extraction Codes/handwritten-tf-1.0-master/readers.py", line 114, in prepare_serialized_examples
features = tf.parse_example(serialized_examples, features=feature_map)
File "/home/nikhil/anaconda/envs/tfone/lib/python2.7/site-packages/tensorflow/python/ops/parsing_ops.py", line 563, in parse_example
dense_types, dense_defaults, dense_shapes, name)
File "/home/nikhil/anaconda/envs/tfone/lib/python2.7/site-packages/tensorflow/python/ops/parsing_ops.py", line 688, in _parse_example_raw
name=name)
File "/home/nikhil/anaconda/envs/tfone/lib/python2.7/site-packages/tensorflow/python/ops/gen_parsing_ops.py", line 178, in _parse_example
dense_shapes=dense_shapes, name=name)
File "/home/nikhil/anaconda/envs/tfone/lib/python2.7/site-packages/tensorflow/python/framework/op_def_library.py", line 768, in apply_op
op_def=op_def)
File "/home/nikhil/anaconda/envs/tfone/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 2336, in create_op
original_op=self._default_original_op, op_def=op_def)
File "/home/nikhil/anaconda/envs/tfone/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 1228, in init
self._traceback = _extract_stack()

InvalidArgumentError (see above for traceback): Name: , Key: imageInput, Index: 0. Number of float values != expected. Values size: 3456000 but output shape: [3240]

command to train
python train.py --slices 26 --width 12 --stride 3 --Bwidth 90 --train_data_pattern batch/tf-data-40x320-81/handwritten-test-{}.tfrecords --train_dir separable_lstm --test_data_pattern batch/tf-data-40x320-81/handwritten-test-{}.tfrecords --max_steps 6000 --batch_size 20 --beam_size 3 --input_chanels 1 --model MDLSTMCTCModel --base_learning_rate 0.001 --num_readers 2 --export_model_steps 500 --display_step 10 --display_step_lme 100 --start_new_model

Please help

Can this be modified to work on custom data? What parameters should be changed and what convention should the data follow?
I am trying to do something similar to https://www.hyperscience.com/forms Would this be possible?

I am running the following :
python train.py --slices 55 --width 12 --stride 1 --Bwidth 350 --vocabulary_size 29
--height 25 --train_data_pattern ./tf-data/handwritten-test-{}.tfrecords --train_dir models-feds
--test_data_pattern ./tf-data/handwritten-test-{}.tfrecords --max_steps 20 --batch_size 20 --beam_size 1
--input_chanels 1 --start_new_model --rnn_cell LSTM --model LSTMCTCModel --num_epochs 6000

Ouput
FutureWarning: Conversion of the second argument of issubdtype from float to np.floating is deprecated. In future, it will be treated as np.float64 == np.dtype(float).type.
from ._conv import register_converters as _register_converters
INFO:tensorflow:/job:master/task:0: Tensorflow version: 1.1.0.
(8750, '', 25, 350, 1)
[20, 25, 350, 1]
0
[20, 25, None, 1]
INFO:tensorflow:/job:master/task:0: Removing existing train directory.
INFO:tensorflow:/job:master/task:0: Flag 'start_new_model' is set. Building a new model.
INFO:tensorflow:Using batch size of 20 for training.
tf-data/handwritten-test-{}.tfrecords
INFO:tensorflow:Number of training files: 3.
(8750, '', 25, 350, 1)
(8750, '', 25, 350, 1)
INFO:tensorflow:Using batch size of 20 for testing.
tf-data/handwritten-test-{}.tfrecords
INFO:tensorflow:Number of testing files: 3.
(8750, '', 25, 350, 1)
(8750, '********************', 25, 350, 1)
Tensor("Reshape:0", shape=(20, 25, 350, 1), dtype=float32)
[20, 25, 350, 1]
0
[20, 25, None, 1]
INFO:tensorflow:/job:master/task:0: Starting managed session.
2018-03-19 18:25:13.630111: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.1 instructions, but these are available on your machine and could speed up CPU computations.
2018-03-19 18:25:13.630151: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.2 instructions, but these are available on your machine and could speed up CPU computations.
2018-03-19 18:25:13.630159: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.
2018-03-19 18:25:13.630174: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2instructions, but these are available on your machine and could speed up CPU computations.
2018-03-19 18:25:13.630180: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.
2018-03-19 18:25:16.371931: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:901] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2018-03-19 18:25:16.372284: I tensorflow/core/common_runtime/gpu/gpu_device.cc:887] Found device 0 with properties:
name: Tesla K80
major: 3 minor: 7 memoryClockRate (GHz) 0.8235
pciBusID 0000:00:1e.0
Total memory: 11.17GiB
Free memory: 11.10GiB
2018-03-19 18:25:16.372318: I tensorflow/core/common_runtime/gpu/gpu_device.cc:908] DMA: 0
2018-03-19 18:25:16.372332: I tensorflow/core/common_runtime/gpu/gpu_device.cc:918] 0: Y
2018-03-19 18:25:16.372347: I tensorflow/core/common_runtime/gpu/gpu_device.cc:977] Creating TensorFlow device (/gpu:0) -> (device: 0, name: Tesla K80, pci bus id: 0000:00:1e.0)
INFO:tensorflow:/job:master/task:0: Entering training loop.
INFO:tensorflow:global_step/sec: 0
INFO:tensorflow:models-feds/model.ckpt-0 is not in all_model_checkpoint_paths. Manually adding it.
2018-03-19 18:25:19.788531: W tensorflow/core/kernels/queue_base.cc:302] _3_test_input/shuffle_batch_join/random_shuffle_queue: Skipping cancelled dequeue attempt with queue not closed
2018-03-19 18:25:19.789465: W tensorflow/core/kernels/queue_base.cc:302] _3_test_input/shuffle_batch_join/random_shuffle_queue: Skipping cancelled dequeue attempt with queue not closed
INFO:tensorflow:/job:master/task:0: Done training -- epoch limit reached.
INFO:tensorflow:/job:master/task:0: Exited training loop.

When I tried running the given commad I got following on the screen...Its not running and repeating the same

INFO:tensorflow:global_step/sec: 0
INFO:tensorflow:separable_lstm/model.ckpt-0 is not in all_model_checkpoint_paths. Manually adding it.
INFO:tensorflow:global_step/sec: 0
INFO:tensorflow:global_step/sec: 0
INFO:tensorflow:global_step/sec: 0
INFO:tensorflow:global_step/sec: 0
INFO:tensorflow:global_step/sec: 0
INFO:tensorflow:global_step/sec: 0
INFO:tensorflow:global_step/sec: 0
INFO:tensorflow:separable_lstm/model.ckpt-0 is not in all_model_checkpoint_paths. Manually adding it.
INFO:tensorflow:global_step/sec: 0

What does vocabulary.txt contain? Do they contain encoded words? Do they contain encoded characters? How was that generated?

Hello there.
I'm getting 'Nonetype object has no attribute prepare_serialised_example's in export_model.py
What do you think is the problem?

I tried to execute the command for training but it is unable to find the tfrecord files needed

NewRandomAccessFile failed to Create/Open: ../tf-data/handwritten-test-54.tfrecords : The system cannot find the path specified.

How do I make the tfrecord files needed for training and testing? :O

Recently I was doing Chinese handwriting recognition, but my crnn method was very bad.Hope you guys give me some suggestions

Peace be upon you,
@johnsmithm

• What are the required dependencies/prerequisite?

• is there nhidden option for training, LSTM units in hidden layer?

• How to train .txt/png pairs?
  example data: Arabic example data.zip

• How to use the trained model to ocr some images?

• Can your software conduct line/layout segmentation & binarization?

I created tfrecords file and tried to run this below command.

python train.py --slices 26 --width 12 --stride 3 --Bwidth 90 --train_data_pattern batch/tf-data-40x320-81/handwritten-test-{}.tfrecords --train_dir separable_lstm --test_data_pattern batch/tf-data-40x320-81/handwritten-test-{}.tfrecords --max_steps 6000 --batch_size 20 --beam_size 3 --input_chanels 1 --model MDLSTMCTCModel --base_learning_rate 0.001 --num_readers 2 --export_model_steps 500 --display_step 10 --display_step_lme 100 --start_new_model

I get this below output:: simply throws exception as epoch limit reached

name: GeForce 940MX
major: 5 minor: 0 memoryClockRate (GHz) 1.2415
pciBusID 0000:01:00.0
Total memory: 1.96GiB
Free memory: 1.59GiB
2018-01-04 18:22:40.984476: I tensorflow/core/common_runtime/gpu/gpu_device.cc:908] DMA: 0
2018-01-04 18:22:40.984483: I tensorflow/core/common_runtime/gpu/gpu_device.cc:918] 0: Y
2018-01-04 18:22:40.984492: I tensorflow/core/common_runtime/gpu/gpu_device.cc:977] Creating TensorFlow device (/gpu:0) -> (device: 0, name: GeForce 940MX, pci bus id: 0000:01:00.0)
INFO:tensorflow:/job:master/task:0: Entering training loop.
INFO:tensorflow:global_step/sec: 0
INFO:tensorflow:separable_lstm/model.ckpt-0 is not in all_model_checkpoint_paths. Manually adding it.
2018-01-04 18:22:42.531717: W tensorflow/core/kernels/queue_base.cc:302] _3_test_input/shuffle_batch_join/random_shuffle_queue: Skipping cancelled dequeue attempt with queue not closed
2018-01-04 18:22:42.532438: W tensorflow/core/kernels/queue_base.cc:302] _3_test_input/shuffle_batch_join/random_shuffle_queue: Skipping cancelled dequeue attempt with queue not closed
INFO:tensorflow:/job:master/task:0: Done training -- epoch limit reached.
INFO:tensorflow:/job:master/task:0: Exited training loop.
(py27_tensorflow_gpu1.1) mlx@mlx-icr:~/Tensorflow_codes/tensorflow_intellichars/handwritten-tf-1.0-master$

Please help

I ran makeTFRecords.ipynb to generate the TFRecords needed for training after the modification made to it in this commit . But it gives me errors in this line:

for ii in range(batchSize): if totalI[ii].shape[0] < 50 or totalI[ii].shape[0] > 200 or \ totalI[ii].shape[1]<10 or totalI[ii].shape[1]>40: print('error!!') continue

I downloaded the handwriting images and xmls needed to make the TFRecords from IAM. I'm using Tensoflow 1.2.1 and the OS of my machine is Windows.

I registered with it and when downloading the dataset, it is throwing error. So sad, any help?

I see you have changed the mdlstm to lstm2d, but I think the lstm2d implemented by tensorflow does not consider the both dimension as introduced by Alex Graves. Have you tested its accuracy?