OpenCV(3.4.3) /io/opencv/modules/imgproc/src/color.cpp:181: error: (-215:Assertion failed) !_src.empty() in function 'cvtColor' about video2tfrecord

[danielbraun89]

I have the same error:

cv2.error("OpenCV(3.4.4) C:\\projects\\opencv-python\\opencv\\modules\\imgproc\\src\\color.cpp:181: error: (-215:Assertion failed) !_src.empty() in function 'cv::cvtColor'\n")

[danielbraun89]

The program crashes even after installing the exact opencv version specified in the requirements.txt file (opencv-python==3.4.0.12),

the error code however, changed to:

OpenCV Error: Assertion failed (scn == 3 || scn == 4) in cv::cvtColor, file C:\projects\opencv-python\opencv\modules\imgproc\src\color.cpp, line 11079

[fascinet]

@ferreirafabio Have you any idea how to resolve this error or any other way for the conversion?

[danielbraun89]

I went ahead and debugged the code, here are my conclusions:

video2tfrecord.py:291

frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
No check is being done on whether the frame is None.
This is causing the original crash

video2tfrecord.py:294

stop, cap, steps, prev_frame_none, frames_counter = repeat_image_retrieval( cap, file_path, take_all_frames, steps, frame, prev_frame_none, frames_counter)

This was probably written for older opencv signature, the new one require the video as a parameter, (and henceforth returns tuple of size 6 and not 5)

No break check or frames_counter increase at

video2tfrecord.py: 301

Causing an infinite loop...

video2tfrecord.py: 294

Using the frame_counter name as unpack variable is overriding the internal counter and again causing an infinite loop...

video2tfrecord.py: 294

Using the cap name as unpack variable and overriding the original cap.
If the cap is being set to None the program again crashes at

video2tfrecord.py: 338

This may be a bug, and maybe not. Im not sure what is the appropriate behavior but in

video2tfrecord.py: 301

It seems video[frames_counter, :, :, :].fill(0) would be more logical than to erase the whole video.

@fascinet I have pushed this changes into a fork: https://github.com/danielbraun89/video2tfrecord

for a quick fix you can just use your own pip installed version and copy the content of https://github.com/danielbraun89/video2tfrecord/blob/master/video2tfrecord.py
into the video2tfrecord.py file that is being used in your installed package.

also I have tested it in the latest opencv, I recommend running a
pip install opencv-python -U
and disregard the requirements.txt opencv version until the official fix will arrive

[fascinet]

Thank you for your reply. It worked very well BTW Great Observation.

[ferreirafabio]

Thank you for these findings and for reporting them. Would you be interested in creating a pull request such that all users would benefit from your bugfixes? I unfortunately don't have time to incorporate these in the next weeks/months. I also don't have time to check whether these reported findings are actually bugs

[ferreirafabio]

Thank you all for your support. The suggestion have now been merged into the master.