shinjayne / shintb Goto Github PK

View Code? Open in Web Editor NEW

90.0 90.0 35.0 112.7 MB

Textboxes : Image Text Detection Model : python package (tensorflow)

Home Page: https://shinjayne.github.io/deeplearning/2017/07/21/text-boxes-paper-review-1.html

Python 100.00%

cnn deep-learning detection ssd tensorflow text-detection textboxes

shintb's People

Contributors

Stargazers

Watchers

shintb's Issues

Trying to access flag --mode before flags were parsed.

Hi ！I am learnig tensorflow and textboxes, when I run the command "python3 main.py --mode train
", it outputs

Traceback (most recent call last): File "main.py", line 24, in <module> if FLAGS.mode == "train": File "/usr/local/lib/python3.5/dist-packages/absl/flags/_flagvalues.py", line 488, in __getattr__ raise _exceptions.UnparsedFlagAccessError(error_message) absl.flags._exceptions.UnparsedFlagAccessError: Trying to access flag --mode before flags were parsed.
How can I run the main.py? Thanks!

It seemed that "h" should be like this "h = float(rectangle.get('height')) / imageHeight"???

shinTB/shintb/svt_data_loader.py

Line 35 in c6c4681

h = float(rectangle.get('height')) / 300.0

		imageWidth = float(image.find('Resolution').get('x'))
		imageHeight = float(image.find('Resolution').get('y'))
		taggedRectangles = image.find('taggedRectangles')
		for rectangle in taggedRectangles.findall('taggedRectangle'):
			h = float(rectangle.get('height')) / imageHeight
			w = float(rectangle.get('width'))  / imageWidth
			x = float(rectangle.get('x'))      / imageWidth
			y = float(rectangle.get('y'))      / imageHeight

Loss goes to NaN during training

After a few hundred global steps, the loss goes to NaN. Any ideas as to why?

how to runner.image?

1.when I run main.py --mode = image occur this problem,how to solve this? If you know, please call me. Thank you @shinjayne .

OpenCV(3.4.1) Error: Assertion failed (depth == 0 || depth == 2 || depth == 5) in cvtColor, file /io/opencv/modules/imgproc/src/color.cpp, line 11109
Traceback (most recent call last):
File "main.py", line 34, in
runner.image()
File "/home/dell/gen/shinTB/shintb/runner.py", line 260, in image
self.outputdrawer.draw_outputs(test_img, output_boxes , output_confidence , wait=1)
File "/home/dell/gen/shinTB/shintb/output_drawer.py", line 100, in draw_outputs
I = cv2.cvtColor(I, cv2.COLOR_RGB2BGR)
cv2.error: OpenCV(3.4.1) /io/opencv/modules/imgproc/src/color.cpp:11109: error: (-215) depth == 0 || depth == 2 || depth == 5 in function cvtColor

2.when I run "runner.image" occur this problem,how to solve this? If you know, please call me. Thank you @shinjayne .

Textboxes information!
rect_start : (0, 0) // rect_end : (0, 0)
confidence: 0.499774
Textboxes information!
rect_start : (0, 0) // rect_end : (0, 0)
confidence: 0.499774
Textboxes information!
rect_start : (0, 0) // rect_end : (0, 0)
confidence: 0.499774
Textboxes information!
rect_start : (0, 0) // rect_end : (0, 0)
confidence: 0.499774
Textboxes information!
rect_start : (0, 0) // rect_end : (0, 0)
confidence: 0.499774
Textboxes information!
rect_start : (0, 0) // rect_end : (0, 0)
confidence: 0.499774
Textboxes information!
rect_start : (0, 0) // rect_end : (0, 0)
confidence: 0.499774
Textboxes information!
rect_start : (0, 0) // rect_end : (0, 0)
confidence: 0.499774

I'm just wondering...

Is there anybody who succeed in running this code normally? because...

There are some issues regrading test / image function.
From the loss curve image, it seems to diverge, not converge.

I trained on tensorflow 1.4, and did so many tries on hyperparameter tuning (learning rate, batch size, batch normalization, etc...)

But all of my effort did not work well.

IF someone who succeed in training this model, can you upload or commit some trained checkpoints?
OR, maybe I should write all of code from scratch.

I use ' runner.test' and I get an error, can you tell me how to solve to solve it?

I run the "runner.test" and I get the error like this ;
" cv2.namedWindow("outputs", cv2.WINDOW_NORMAL)
cv2.error: /io/opencv/modules/highgui/src/window.cpp:565: error: (-2) The function is not implemented. Rebuild the library with Windows, GTK+ 2.x or Carbon support. If you are on Ubuntu or Debian, install libgtk2.0-dev and pkg-config, then re-run cmake or configure script in function cvNamedWindow "
And I delete " cv2.namedWindow("outputs", cv2.WINDOW_NORMAL) " in the runner.test, then I get this;
" File "/home/tian/tensorflow/example/11/shinTB-master/main.py", line 22, in
runner.test(1000)
File "/home/tian/tensorflow/example/11/shinTB-master/shintb/runner.py", line 196, in test
self.outputdrawer.draw_outputs(test_imgs[0], output_boxes, output_confidence, wait=1)
File "/home/tian/tensorflow/example/11/shinTB-master/shintb/output_drawer.py", line 104, in draw_outputs
I = cv2.rectangle(I, rect_start, rect_end, (255, 0, 0) , 5 )
TypeError: integer argument expected, got float "

And the returned value is :
" FILTERED BOXES INFO : [([-inf, -inf, inf, inf], 0.49975014, 0)...
PICKED BOXES INFO : [([-inf, -inf, inf, inf], 0.49975014, 0)] "

can you tell me how to solve this problem?
Thank you for your help.

Anyone has trained the model correctly?

I trained the model with loss is NAN.
I couldn't figure out whether the problem laid in structure or in training strategey.

question about runner.test()

Hi, Thank you for great code!!, I am newbie in Textboxes.
After training runner.train(..), I have evaluated runner.test(2) and then I have error like:

TypeError Traceback (most recent call last)
in ()
----> 1 runner.test(2)

E:\shinTB\shintb\runner.py in test(self, iter)
194 output_boxes, output_confidence = self.outputdrawer.format_output(pred_conf[0], pred_loc[0])
195
--> 196 self.outputdrawer.draw_outputs(test_imgs[0], output_boxes, output_confidence, wait=1)
197
198 step += 1

E:\shinTB\shintb\output_drawer.py in draw_outputs(self, img, boxes, confidences, wait)
102 rect_start = (x,y)
103 rect_end = (x+w, y+h)
--> 104 I = cv2.rectangle(I, rect_start, rect_end, (255, 0, 0) , 5 )
105
106 print("Textboxes information!")

TypeError: integer argument expected, got float

Can you fix it?

Training model

Do you mind giving me the training model for me (ckpt file)?

The errors when I did test.

The following is code that I used for testing the model after training.
from config import config
from shintb import graph_drawer, default_box_control, svt_data_loader, runner,output_drawer
import tensorflow as tf
flags = tf.app.flags
FLAGS = flags.FLAGS
graphdrawer = graph_drawer.GraphDrawer(config)
dataloader = svt_data_loader.SVTDataLoader('./svt2/train.xml','./svt2/test.xml')
dbcontrol = default_box_control.DefaultBoxControl(config, graphdrawer)
outputdrawer = output_drawer.OutputDrawer(config, dbcontrol)
runner = runner.Runner(config, graphdrawer, dataloader, dbcontrol,outputdrawer)
flags.DEFINE_integer("iter", 10000, "iteration for job")
runner.test(FLAGS.iter)

But I have the following errors:
/home/jsj/shinTB/shintb/utils/box_calculation.py:52: RuntimeWarning: invalid value encountered in double_scalars
right = min(r1[0] + r1[2], r2[0] + r2[2])
/home/jsj/shinTB/shintb/utils/box_calculation.py:53: RuntimeWarning: invalid value encountered in double_scalars
bottom = min(r1[1] + r1[3], r2[1] + r2[3])
PICKED BOXES INFO : [([-inf, 4260720.561588916, inf, 0.044721359549995794], 0.49975014, 0)]
/home/jsj/shinTB/shintb/output_drawer.py:103: RuntimeWarning: invalid value encountered in double_scalars
rect_end = (x+w, y+h)
Traceback (most recent call last):
File "/home/jsj/shinTB/new_file_for_test.py", line 12, in
runner.test(FLAGS.iter)
File "/home/jsj/shinTB/shintb/runner.py", line 196, in test
self.outputdrawer.draw_outputs(test_imgs[0], output_boxes, output_confidence, wait=1)
File "/home/jsj/shinTB/shintb/output_drawer.py", line 104, in draw_outputs
I = cv2.rectangle(I, rect_start, rect_end, (255, 0, 0) , 5 )
TypeError: integer argument expected, got float.

Can you help me fix the problem?
@shinjayne

Couldn't find good training data

Model not correct

I tried to train and test the model but I think that there are many and many errors within it. The most important is that confidence and delta are predicted as really big numbers (i.e. 1.8e+8). This make the project not usable until a real deep analysis of it

where are you rnn layer?

I cant find lstm layer. you code it is seems like ssd-vgg16, but not like textboxes

missing 1 required positional argument: 'outputdrawer'

Getting this error when I run "runner = runner.Runner(config, graphdrawer, dataloader, dbcontrol)" :-
Traceback (most recent call last):
File "", line 1, in
TypeError: init() missing 1 required positional argument: 'outputdrawer'

Error happens while saving checkpoint. How to fix it? Thanks a lot

GLOBAL STEP : 1000 / LEARNING RATE : 0.0008 / LOSS : 1.18235 ( 0.27370095253 secs)
Traceback (most recent call last):
File "main.py", line 30, in
runner.train(FLAGS.jobname, FLAGS.iter)
File "/home/wanghz/shinTB/shintb/runner.py", line 101, in train
ckpt_path = self.saver.save(self.sess, "%s.ckpt" % (c["saved_dir"]+"/"+jobname),global_step)
File "/opt/DL/tensorflow/lib/python2.7/site-packages/tensorflow/python/training/saver.py", line 1369, in save
self.last_checkpoints, latest_filename)
File "/opt/DL/tensorflow/lib/python2.7/site-packages/tensorflow/python/training/saver.py", line 796, in update_checkpoint_state
text_format.MessageToString(ckpt))
File "/opt/DL/tensorflow/lib/python2.7/site-packages/tensorflow/python/lib/io/file_io.py", line 353, in atomic_write_string_to_file
rename(temp_pathname, filename, overwrite=True)
File "/opt/DL/tensorflow/lib/python2.7/site-packages/tensorflow/python/lib/io/file_io.py", line 335, in rename
compat.as_bytes(oldname), compat.as_bytes(newname), overwrite, status)
File "/usr/lib/python2.7/contextlib.py", line 24, in exit
self.gen.next()
File "/opt/DL/tensorflow/lib/python2.7/site-packages/tensorflow/python/framework/errors_impl.py", line 466, in raise_exception_on_not_ok_status
pywrap_tensorflow.TF_GetCode(status))
tensorflow.python.framework.errors_impl.FailedPreconditionError: saved/checkpoint.tmp205d246fabab4cb488968fcc6989f6d7

line 251 in default_box_control.py should be sum_exp_a = exp_a.sum(reduce_axis) ???

shinTB/shintb/default_box_control.py

Line 251 in c6c4681

sum_exp_a = a.sum(reduce_axis)

Module output_drawer has no attribute 'formal_output'

Has anyone encountered this error when tryint to "test" or "image"?

Traceback (most recent call last):
  File "testShinTB.py", line 15, in <module>
    runner.test(5)
  File "/home/great/Documentos/spi/shinTB/shintb/runner.py", line 194, in test
    output_boxes, output_confidence = self.outputdrawer.format_output(pred_conf[0], pred_loc[0])
AttributeError: module 'shintb.output_drawer' has no attribute 'format_output'

I am just tryint to run an example based on the README.

Some questions about code, HELP!

@shinjayne
Thanks a lot for your implementation and contribution!

First of all, I like the way you write program. I spent couple days to read your programs. However, there are some questions that I could not understand after deliberation.

Here is the question

line 55-56 in default_box_control.py, I print all information of scale, w, and h. however, what's the really meaning of default_w and default_h, according to line 57,58, where c_x, and c_y is normalized between 0 and 1. Is it possible that w and h will be normalized with respect to width and height of the original image?
default_w = scale * np.sqrt(rs[i])
default_h = scale / np.sqrt(rs[i])

scale: 0.10, box_ratio: 1.00, w: 0.10 h: 0.10
scale: 0.10, box_ratio: 2.00, w: 0.14 h: 0.07
scale: 0.10, box_ratio: 3.00, w: 0.17 h: 0.06
scale: 0.10, box_ratio: 5.00, w: 0.22 h: 0.04
scale: 0.10, box_ratio: 7.00, w: 0.26 h: 0.04
scale: 0.10, box_ratio: 10.00, w: 0.32 h: 0.03
scale: 0.27, box_ratio: 1.00, w: 0.27 h: 0.27
scale: 0.27, box_ratio: 2.00, w: 0.38 h: 0.19
scale: 0.27, box_ratio: 3.00, w: 0.47 h: 0.16
scale: 0.27, box_ratio: 5.00, w: 0.60 h: 0.12
scale: 0.27, box_ratio: 7.00, w: 0.71 h: 0.10
scale: 0.27, box_ratio: 10.00, w: 0.85 h: 0.09
scale: 0.44, box_ratio: 1.00, w: 0.44 h: 0.44
scale: 0.44, box_ratio: 2.00, w: 0.62 h: 0.31
scale: 0.44, box_ratio: 3.00, w: 0.76 h: 0.25
scale: 0.44, box_ratio: 5.00, w: 0.98 h: 0.20
scale: 0.44, box_ratio: 7.00, w: 1.16 h: 0.17
scale: 0.44, box_ratio: 10.00, w: 1.39 h: 0.14
scale: 0.61, box_ratio: 1.00, w: 0.61 h: 0.61
scale: 0.61, box_ratio: 2.00, w: 0.86 h: 0.43
scale: 0.61, box_ratio: 3.00, w: 1.06 h: 0.35
scale: 0.61, box_ratio: 5.00, w: 1.36 h: 0.27
scale: 0.61, box_ratio: 7.00, w: 1.61 h: 0.23
scale: 0.61, box_ratio: 10.00, w: 1.93 h: 0.19
scale: 0.78, box_ratio: 1.00, w: 0.78 h: 0.78
scale: 0.78, box_ratio: 2.00, w: 1.10 h: 0.55
scale: 0.78, box_ratio: 3.00, w: 1.35 h: 0.45
scale: 0.78, box_ratio: 5.00, w: 1.74 h: 0.35
scale: 0.78, box_ratio: 7.00, w: 2.06 h: 0.29
scale: 0.78, box_ratio: 10.00, w: 2.47 h: 0.25
scale: 0.95, box_ratio: 1.00, w: 0.95 h: 0.95
scale: 0.95, box_ratio: 2.00, w: 1.34 h: 0.67
scale: 0.95, box_ratio: 3.00, w: 1.65 h: 0.55
scale: 0.95, box_ratio: 5.00, w: 2.12 h: 0.42
scale: 0.95, box_ratio: 7.00, w: 2.51 h: 0.36
scale: 0.95, box_ratio: 10.00, w: 3.00 h: 0.30

line 35-38 in svt_data_loader.py, why h is divided by 300 instead of the height and width of the original image?
h = float(rectangle.get('height')) / 300.0
w = float(rectangle.get('width')) / 300.0
x = float(rectangle.get('x')) / 300.0
y = float(rectangle.get('y')) / 300.0

BTW, how many epochs will lead to converge based on the given training data?

Looking forward to Your reply. Thanks a lot.

shinjayne / shintb Goto Github PK

shintb's People

Contributors

Stargazers

Watchers

Forkers

shintb's Issues

Recommend Projects

Recommend Topics

Recommend Org