make error: arithmetic on a pointer to void about darknet

[prabindh]

You could potentially refer to the latest code from the branch darknet-cpp-windows, as it is fixed there (https://github.com/prabindh/darknet/tree/darknet-cpp-windows/src/utils.c). I will pull in the fix to the master after some more productising of the windows branch. Thanks for noting this.

[prabindh]

Could you also update the gcc version being used on your side ?

[eisneim]

@prabindh

$ gcc --version
Configured with: --prefix=/Applications/Xcode.app/Contents/Developer/usr --with-gxx-include-dir=/usr/include/c++/4.2.1
Apple LLVM version 8.0.0 (clang-800.0.42.1)
Target: x86_64-apple-darwin16.3.0

[eisneim]

@prabindh the window version's code fixed the issue 👍

[eisneim]

@prabindh just a quick question, i was using original and your fork of darknet-cpp to train my custom datasets(7 classes, 500x500resolution), they all fails randomly with segmentatiion fault after random iterations

[1]    31987 segmentation fault  ./darknet-cpp detector train fsd/fsd.data fsd/yolo-fsd.cfg

I'm not sure whether this is caused by my image files, so i use opencv to convert all my images to jpg, then i try to put tons of printf() in source code to see which step messed up the memory, but still no luck. maybe this time i'll try to use photoshop to batch conversion

any suggestions?
thanks

[prabindh]

Are the label files following Yolo format?  Sent from Yahoo Mail on Android On Thu, Mar 9, 2017 at 6:30 PM, Eisneim Terry<[email protected]> wrote: @prabindh just a quick question, i was using original and your fork of darknet-cpp to train my custom datasets(7 classes, 500x500resolution), they all fails randomly with segmentatiion fault after random iterations [1] 31987 segmentation fault ./darknet-cpp detector train fsd/fsd.data fsd/yolo-fsd.cfg I'm not sure whether this is caused by my image files, so i use opencv to convert all my images to jpg, then i try to put tons of printf() in source code to see which step messed up the memory, but still no luck. maybe this time i'll try to use photoshop to batch conversion any suggestions? thanks — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

[eisneim]

it just failed randomly, i logged out the image and label just before it fails, but i can see no issue.

i wrote a chrome extension for creating annotation: https://github.com/eisneim/dl_annotator scripts folder contains my scripts to generate yolo labels,
each label:

2 0.265196 0.137720 0.570056 0.515504

my scripts ensures each rect box's x,y value must be greater than 0 and less than 1.0
maybe i should check it's order, minX, minY always first

anyway, thanks for your reply

[prabindh]

After looking at https://github.com/eisneim/dl_annotator/blob/master/scripts/yolo_darknet.py, specifically

x = min(x1, x2) / width

It appears that the labelling is incorrect. Please check https://github.com/prabindh/euclid/blob/master/euclid.py for the coordinate format for the bounding boxes of Yolo.

[eisneim]

@prabindh oh my! i just realized x y are center of rectangle! thank you

[prabindh]

Please do not forget to close the issue if it is fixed. Thanks. You can raise a separate issue, for the crash that happens if labels are invalid.

[eisneim]

thanks for your help, i got 30fps results

[prabindh]

Just to let you know, the windows port has now been merged into master. Thx