0 prediction with arapaho build (Yolo V3)

[turkishviking]

git log : b177eed22ad3ab7335d2171b12e2a91e6829fbff GPU=1 CUDNN=1 OPENCV=1

Linux Mint 18.3 (~Ubuntu 16.04) OpenCV: 3.4.0 Cudnn 5.1 Cuda 8

Hi Prabindh, I'm testing the V3 version, when using 'make darknet-cpp', with weights files from https://pjreddie.com/darknet/yolo/ (yoloV3-608) all is fine. But when I try with the same configuration to 'make arapaho' (with libdarknet-cpp-shared.so pre-build with same configuration too), I have 0 prediction (and no crash / error log) Any idea of what I could do wrong?

Edit:

• I have tried all this afternoon, with the same input.weight and input cfg the two version (darknet-cpp and arapaho), the darknet-cpp works, the arapaho doesn't detect anything, can it be related to the input.data ?

• tried to compile darknet-cpp-shared without opencv, --> detections == 0

• Image seems valid, I can see it in cv::NamedWindow, and print it: Image data = 0x7fe3a586f040, w = 768, h = 576

• Also checked layers:

  
  net = `parse_network_cfg(cfgfile);`

layer l0 = net.layers[0]; cout << l0.w << endl; layer l1 = net.layers[1]; cout << l1.w << endl; layer l = net.layers[net.n - 1]; cout << l.w << endl;

Which give me

416 416 52

[prabindh]

Please check latest master. There were changes to be done for iterating through the detections.

[turkishviking]

Ok, I was modifying exactly the same things to make a pull request, cool, that's work, thank's!