regarding training accuracy and detection

[Ram-Godavarthi]

First of all the training on CPU is very slow.

training script is running properly. But I am not able to get the correct detections at all.

I am getting something like 👍

b'NRRRRRRRRRRRRRRRRRRRR'

b'M99999RRRRRRRRRRRRRRR'

Something like this..

Where do I need to the give the pretrained models and start detection process??

Could you please help me out with this??

Thank you

[codeaway23]

Before running the eval script you will have to add the path to your training logs and checkpoint files in the common_flags.py file in the models/research/attention_ocr/python directory

[Ram-Godavarthi]

I am getting like this when I run the training script.

INFO:tensorflow:global step 176: loss = 45.1903 (4.068 sec/step)
INFO 2020-01-28 10:31:35.000737: tf_logging.py: 82 global step 176: loss = 45.1903 (4.068 sec/step)
INFO:tensorflow:global step 177: loss = 44.7751 (4.017 sec/step)
INFO 2020-01-28 10:31:39.000878: tf_logging.py: 82 global step 177: loss = 44.7751 (4.017 sec/step)
INFO:tensorflow:Recording summary at step 177.
INFO 2020-01-28 10:31:44.000128: tf_logging.py: 82 Recording summary at step 177.
INFO:tensorflow:global step 178: loss = 47.6715 (8.347 sec/step)
INFO 2020-01-28 10:31:48.000249: tf_logging.py: 82 global step 178: loss = 47.6715 (8.347 sec/step)
INFO:tensorflow:global step 179: loss = 57.1654 (4.037 sec/step)
INFO 2020-01-28 10:31:52.000310: tf_logging.py: 82 global step 179: loss = 57.1654 (4.037 sec/step)
INFO:tensorflow:global step 180: loss = 44.0419 (4.018 sec/step)
INFO 2020-01-28 10:31:56.000362: tf_logging.py: 82 global step 180: loss = 44.0419 (4.018 sec/step)
INFO:tensorflow:global step 181: loss = 47.7864 (4.122 sec/step)
INFO 2020-01-28 10:32:00.000507: tf_logging.py: 82 global step 181: loss = 47.7864 (4.122 sec/step)
INFO:tensorflow:global step 182: loss = 46.3507 (4.042 sec/step)
INFO 2020-01-28 10:32:04.000572: tf_logging.py: 82 global step 182: loss = 46.3507 (4.042 sec/step)
INFO:tensorflow:global step 183: loss = 46.4693 (4.069 sec/step)
INFO 2020-01-28 10:32:08.000677: tf_logging.py: 82 global step 183: loss = 46.4693 (4.069 sec/step)
INFO:tensorflow:global step 184: loss = 48.3467 (4.057 sec/step)
INFO 2020-01-28 10:32:12.000758: tf_logging.py: 82 global step 184: loss = 48.3467 (4.057 sec/step)
INFO:tensorflow:global step 185: loss = 42.2563 (4.047 sec/step)
INFO 2020-01-28 10:32:16.000839: tf_logging.py: 82 global step 185: loss = 42.2563 (4.047 sec/step)
INFO:tensorflow:global step 186: loss = 46.8351 (4.087 sec/step)
INFO 2020-01-28 10:32:20.000948: tf_logging.py: 82 global step 186: loss = 46.8351 (4.087 sec/step)
INFO:tensorflow:global step 187: loss = 44.3799 (4.096 sec/step)
INFO 2020-01-28 10:32:25.000081: tf_logging.py: 82 global step 187: loss = 44.3799 (4.096 sec/step)
INFO:tensorflow:global step 188: loss = 45.6797 (4.065 sec/step)
INFO 2020-01-28 10:32:29.000168: tf_logging.py: 82 global step 188: loss = 45.6797 (4.065 sec/step)
INFO:tensorflow:global step 189: loss = 47.7265 (4.045 sec/step)
INFO 2020-01-28 10:32:33.000237: tf_logging.py: 82 global step 189: loss = 47.7265 (4.045 sec/step)
INFO:tensorflow:global step 190: loss = 45.3303 (4.070 sec/step)
INFO 2020-01-28 10:32:37.000429: tf_logging.py: 82 global step 190: loss = 45.3303 (4.070 sec/step)
INFO:tensorflow:Recording summary at step 190.
INFO 2020-01-28 10:32:41.000687: tf_logging.py: 82 Recording summary at step 190.
INFO:tensorflow:global step 191: loss = 46.0952 (8.345 sec/step)
INFO 2020-01-28 10:32:45.000797: tf_logging.py: 82 global step 191: loss = 46.0952 (8.345 sec/step)
INFO:tensorflow:global step 192: loss = 36.7755 (4.066 sec/step)
INFO 2020-01-28 10:32:49.000887: tf_logging.py: 82 global step 192: loss = 36.7755 (4.066 sec/step)
INFO:tensorflow:global step 193: loss = 46.4757 (4.032 sec/step)
INFO 2020-01-28 10:32:53.000941: tf_logging.py: 82 global step 193: loss = 46.4757 (4.032 sec/step)
INFO:tensorflow:global step 194: loss = 51.2351 (4.016 sec/step)
INFO 2020-01-28 10:32:57.000992: tf_logging.py: 82 global step 194: loss = 51.2351 (4.016 sec/step)
INFO:tensorflow:global step 195: loss = 51.1448 (4.074 sec/step)
INFO 2020-01-28 10:33:02.000088: tf_logging.py: 82 global step 195: loss = 51.1448 (4.074 sec/step)
INFO:tensorflow:global step 196: loss = 47.8617 (4.045 sec/step)
INFO 2020-01-28 10:33:06.000155: tf_logging.py: 82 global step 196: loss = 47.8617 (4.045 sec/step)
INFO:tensorflow:global step 197: loss = 31.9874 (4.075 sec/step)
INFO 2020-01-28 10:33:10.000264: tf_logging.py: 82 global step 197: loss = 31.9874 (4.075 sec/step)
INFO:tensorflow:global step 198: loss = 45.3187 (4.025 sec/step)
INFO 2020-01-28 10:33:14.000314: tf_logging.py: 82 global step 198: loss = 45.3187 (4.025 sec/step)
INFO:tensorflow:global step 199: loss = 49.2523 (4.045 sec/step)
INFO 2020-01-28 10:33:18.000393: tf_logging.py: 82 global step 199: loss = 49.2523 (4.045 sec/step)
INFO:tensorflow:global step 200: loss = 48.7986 (6.094 sec/step)
INFO 2020-01-28 10:33:24.000510: tf_logging.py: 82 global step 200: loss = 48.7986 (6.094 sec/step)
INFO:tensorflow:Stopping Training.
INFO 2020-01-28 10:33:24.000534: tf_logging.py: 82 Stopping Training.
INFO:tensorflow:Finished training! Saving model to disk.
INFO 2020-01-28 10:33:24.000534: tf_logging.py: 82 Finished training! Saving model to disk.
(env) ram@virtualbox:~/attention_ocr/python$

What could be the problem??

[Ram-Godavarthi]

And I got this after running eval.py script.

INFO 2020-01-28 10:46:05.000354: tf_logging.py: 82 Evaluation [99/100]
INFO:tensorflow:Evaluation [100/100]
INFO 2020-01-28 10:46:06.000709: tf_logging.py: 82 Evaluation [100/100]
INFO:tensorflow:Finished evaluation at 2020-01-28-09:46:06
INFO 2020-01-28 10:46:06.000709: tf_logging.py: 82 Finished evaluation at 2020-01-28-09:46:06
2020-01-28 10:46:07.241598: I tensorflow/core/kernels/logging_ops.cc:79] eval/SequenceAccuracy[0]
2020-01-28 10:46:07.241830: I tensorflow/core/kernels/logging_ops.cc:79] eval/CharacterAccuracy[0.0399122499]
INFO:tensorflow:Waiting for new checkpoint at datasets/data/model_checkpoints
INFO 2020-01-28 10:46:08.000799: tf_logging.py: 82 Waiting for new checkpoint at datasets/data/model_checkpoints

Could you tell me where is it going wrong

[Ram-Godavarthi]

When I run demo_inference,py script I get this result.

DEBUG 2020-01-28 11:01:54.000358: PngImagePlugin.py: 138 STREAM b'IHDR' 16 13
DEBUG 2020-01-28 11:01:54.000358: PngImagePlugin.py: 138 STREAM b'IDAT' 41 43382
datasets/data/crops/3.png number_plates 1
This part is executing
Reading filepattern/1.png
DEBUG 2020-01-28 11:01:54.000369: PngImagePlugin.py: 138 STREAM b'IHDR' 16 13
DEBUG 2020-01-28 11:01:54.000369: PngImagePlugin.py: 138 STREAM b'IDAT' 41 43382
datasets/data/crops/3.png number_plates 2
This part is executing
Reading filepattern/2.png
DEBUG 2020-01-28 11:01:54.000377: PngImagePlugin.py: 138 STREAM b'IHDR' 16 13
DEBUG 2020-01-28 11:01:54.000377: PngImagePlugin.py: 138 STREAM b'IDAT' 41 43382
datasets/data/crops/3.png number_plates 3
This part is executing
Reading filepattern/3.png
DEBUG 2020-01-28 11:01:54.000380: PngImagePlugin.py: 138 STREAM b'IHDR' 16 13
DEBUG 2020-01-28 11:01:54.000383: PngImagePlugin.py: 138 STREAM b'IDAT' 41 43382
2020-01-28 11:01:54.889808: I tensorflow/core/platform/cpu_feature_guard.cc:137] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2
INFO:tensorflow:Restoring parameters from datasets/data/model_checkpoints/model.ckpt-200
INFO 2020-01-28 11:01:54.000890: tf_logging.py: 82 Restoring parameters from datasets/data/model_checkpoints/model.ckpt-200
b'444444444444444444444'
b'444444444444444444444'
b'444444444444444444444'
b'444444444444444444444'

Please try to help me out in this issue.

What's going wrong

[Rubeen-Mohammad]

Hello! Thank you so much for such nicely explained kernels. I too have the same kind of problem.
I have added the path to training logs and checkpoint files in the common_flags.py file in the models/research/attention_ocr/python directory

Problem1 #####:

Even though a new checkpoint is available, it is unable to collect that checkpoint and it is waiting for a long time. Could you please tell me what's the problem?
My evaluation is getting struck as follows:
INFO 2020-03-23 13:29:03.000567: evaluation.py: 169 Evaluation [90/100]
INFO:tensorflow:Evaluation [100/100]
INFO 2020-03-23 13:29:16.000567: evaluation.py: 169 Evaluation [100/100]
INFO:tensorflow:Finished evaluation at 2020-03-23-07:59:16
INFO 2020-03-23 13:29:16.000567: evaluation.py: 455 Finished evaluation at 2020-03-23-07:59:16
eval/SequenceAccuracy[0]
eval/CharacterAccuracy[0.00575520797]
INFO:tensorflow:Waiting for new checkpoint at /home/mwebware/Documents/Rubeen/Text_Recognition2/models-master/research/attention_ocr/python/aocr-logs
INFO 2020-03-23 13:29:18.000558: evaluation.py: 189 Waiting for new checkpoint at /home/mwebware/Documents/Rubeen/Text_Recognition2/models-master/research/attention_ocr/python/aocr-logs

Problem2 #####:

After running inference.py, my model is predicting the license plate characters as follows:
-----------Predicted------------- [b'5555555\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91', b'555\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91', b'111111\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91', b'111111\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91', b'5555\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91', b'555\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91', b'555\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91', b'555\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91']
b'5555555\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91'
b'555\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91'
b'111111\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91'
b'111111\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91'
b'5555\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91'
b'555\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91'
b'555\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91'
b'555\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91\xe2\x96\x91'
Could you please help me in resolving this.
Thank you

[Ram-Godavarthi]

Problem 1: I just tried 1 time. So, whenever I need it, I ran the script to check the accuracy.

Problem 2: Just add convert the output format by inserting this line.

line.decode('utf-8')