AssertionError: expect 44948596 bytes, found 44948600

[zef1611]

so when i use this command on cmd c:\dark\darkflow-master>python flow --model cfg/yolov2-tiny.cfg --load bin/yolov2-tiny.weights
and then this error appears

please help me fix this

[ideaRunner]

Hi, I have this issue too.
same yolov2-tiny.
AssertionError: expect 63820056 bytes, found 63820060
Can anyone help us?

[ideaRunner]

Oh, Jesus!
Come to see what I have found here!
https://sites.google.com/view/tensorflow-example-java-api/complete-guide-to-train-yolo/convert-darknet-weights-to-pb-file

[zef1611]

thank you for that link
and btw do you want to contact with me so we can help each other in this darkflow stuff?

[Rabiah86]

I had the same error with yolo not with tiny-yolo,
try to reduce the width and height in the cfg file.it solved the problem for me .
keep in mind that width and height values should be a multiple of 32

[derekwong66]

I got the same problem. I modified yolov2.cfg with 3 classes and used yolov2.weights to train my own dataset.
I tried this method https://sites.google.com/view/tensorflow-example-java-api/complete-guide-to-train-yolo/convert-darknet-weights-to-pb-file to solve the problem. However, it doesn't work. I replaced the offset = 16 with offset = 20, even offset = 40.

zihang@derek:~/project/darkflow$ flow --model cfg/yolov2-3c.cfg --load bin/yolov2.weights --train --annotation train/Annotations --dataset train/Images --gpu 1.0

/home/zihang/anaconda3/envs/tensorflow/lib/python3.6/site-packages/darkflow/dark/darknet.py:54: UserWarning: ./cfg/yolov2.cfg not found, use cfg/yolov2-3c.cfg instead
cfg_path, FLAGS.model))
Parsing cfg/yolov2-3c.cfg
Loading bin/yolov2.weights ...
Traceback (most recent call last):
File "/home/zihang/anaconda3/envs/tensorflow/bin/flow", line 6, in
cliHandler(sys.argv)
File "/home/zihang/anaconda3/envs/tensorflow/lib/python3.6/site-packages/darkflow/cli.py", line 26, in cliHandler
tfnet = TFNet(FLAGS)
File "/home/zihang/anaconda3/envs/tensorflow/lib/python3.6/site-packages/darkflow/net/build.py", line 58, in init
darknet = Darknet(FLAGS)
File "/home/zihang/anaconda3/envs/tensorflow/lib/python3.6/site-packages/darkflow/dark/darknet.py", line 27, in init
self.load_weights()
File "/home/zihang/anaconda3/envs/tensorflow/lib/python3.6/site-packages/darkflow/dark/darknet.py", line 82, in load_weights
wgts_loader = loader.create_loader(*args)
File "/home/zihang/anaconda3/envs/tensorflow/lib/python3.6/site-packages/darkflow/utils/loader.py", line 105, in create_loader
return load_type(path, cfg)
File "/home/zihang/anaconda3/envs/tensorflow/lib/python3.6/site-packages/darkflow/utils/loader.py", line 19, in init
self.load(*args)
File "/home/zihang/anaconda3/envs/tensorflow/lib/python3.6/site-packages/darkflow/utils/loader.py", line 77, in load
walker.offset, walker.size)
AssertionError: expect 202355760 bytes, found 203934260

[dudqls1994]

I also same error.... I cannot solved it.....Anyone help me

[carlchandev]

Oh, Jesus!
Come to see what I have found here!
https://sites.google.com/view/tensorflow-example-java-api/complete-guide-to-train-yolo/convert-darknet-weights-to-pb-file

I don't understand why, but it works!!! thank you!!

[seek-ing]

@dudqls1994
1、首先，我们在ubuntu下先用anaconda2（版本是2还是3不影响）安装了tensorflow.
2、darkflow要求的安装环境是Python3，那么把anaconda2里的Python2升级到python3.5就行，Ubuntu系统默认是自带python2.7的，在使用darkflow时，用source activate py35进入Python3.5环境再使用flow命令就行。
3、接下来，你可以在任何地方安装darkflow，但是要在py35环境下安装，换Python环境source activate py35，然后安装好你会看到darkflow的文件夹，这个文件夹没用，在安装darkflow的时候会自动拷贝到anaconda2/envs/py34/lib/python3.5/site-packages/darkflow/utils（我Python3.5命名的时候手抖，弄成了py34），要改的是上述目录下loader.py第121行，把self.offset = 16改成self.offset = 20

[SuperStriker]

Oh, Jesus!
Come to see what I have found here!
https://sites.google.com/view/tensorflow-example-java-api/complete-guide-to-train-yolo/convert-darknet-weights-to-pb-file

Thanks man, you are a life saver

[AbanoubMamdouh]

Tried @ideaRunner solution, didn't completely work, displayed another error, so I reduced the gpu from 1.0 to 0.8 and it worked fine! Thanks @ideaRunner! Sorry for not attaching any screenshots, I would have to recreate the problem.
N.B: If you use the same gpu for your display, it might create a problem to use 0.8 of the gpu, so I recommend you reduce gpu usage to 0.7 and it will work fine!

[danFromTelAviv]

Oh, Jesus!
Come to see what I have found here!
https://sites.google.com/view/tensorflow-example-java-api/complete-guide-to-train-yolo/convert-darknet-weights-to-pb-file

so we don't have to go into the link that may break one day :

The solution is very simple, let's modify the line self.offset = 16 in the ./darkflow/utils/loader.py file and replace with self.offset = 20.

For me it did not work. the difference between expect and found shrank a lot though

[shalin51]

Oh, Jesus!
Come to see what I have found here!
https://sites.google.com/view/tensorflow-example-java-api/complete-guide-to-train-yolo/convert-darknet-weights-to-pb-file

so we don't have to go into the link that may break one day :

The solution is very simple, let's modify the line self.offset = 16 in the ./darkflow/utils/loader.py file and replace with self.offset = 20.

For me it did not work. the difference between expect and found shrank a lot though

I had same problem but it can be solved by correct self.offset

if 44948596 bytes, found 44948600

means the difference (44948600- 44948596 ) should be added to self.offset.

self.offset= self.offset + 4 difference (44948600- 44948596 )

[danFromTelAviv]

face palm*
thanks @shalin51

[mpky]

@shalin51 Does that mean that new walk function should be:

def walk(self, size):
if self.eof: return None
end_point = self.offset + 4 * (44948600- 44948596 )
assert end_point <= self.size,
'Over-read {}'.format(self.path)

Should we just delete the size parameter then?

[chenminhua]

Here is my solution:

1. pip uninstall darkflow
2. change the source code in ./darkflow/utils/loader.py, just delete the "assert statement" or replace it by a warning
3. reinstall it by pip install -e .

It works for me.

Notice: You still have to figure out that whether your cfg file and weights and labels.txt match to each other.

[540ul4rm6]

I solved the problem.

http://tn00343140a.pixnet.net/blog/post/299895872-yolo%e9%81%8b%e8%a1%8cdarkflow%e6%99%82%2c-%e5%87%ba%e7%8f%beassertionerror%3a-expect-20392940

Find the file "loader.py" and change the value of "self.offset".
updated self.offset = old_offset_value + (found_value - expected_value)

[santarone]

I also solved.

im just changing self.offset value as '20' "loader.py" in ~/darkflow/utils

i don't know why this working. but it's work!

[rammanoj]

I've been trying all the above specified methods. But nothing of it seem to be working!!

Is there any other way to solve this issue up!!

[TDA1JU]

Oh, Jesus!
Come to see what I have found here!
https://sites.google.com/view/tensorflow-example-java-api/complete-guide-to-train-yolo/convert-darknet-weights-to-pb-file

Honestly I do not know how does it work, but it solved my issue, great!!

[senkumartup]

diff --git a/darkflow/utils/loader.py b/darkflow/utils/loader.py
index 723560d..e2fdaea 100644
--- a/darkflow/utils/loader.py
+++ b/darkflow/utils/loader.py
@@ -118,7 +118,8 @@ class weights_walker(object):
shape = (), mode = 'r', offset = 0,
dtype = '({})i4,'.format(4))
self.transpose = major > 1000 or minor > 1000

•        self.offset = 16
  

•        self.offset = 20
  

  def walk(self, size):
  if self.eof: return None