Michael Nielsen

This is still a problem, since we cant access the pretrainednetworks . (cntk-py36) C:\local\Anaconda3-4.1.1-Windows-x86_64\envs\cntk-py36\Lib\site-packages\cntk\contrib\crosstalkcaffe\examples>python run_convert.py -n ResNet152 Warning: Un-supported bvlccaffe type: scale_conv1-Scale Warning: Un-supported bvlccaffe type: scale2a_branch1-Scale Warning: Un-supported bvlccaffe...

The output I posted above was only one type of error I got trying to convert. There is another error type that I could fix, where it stoped on attribute...

Yep, it works now :) Meanwhile I converted some nice caffe models yesterday: - googlenet age and gender - googlenet imagenet - vgg16 pascal (classification) and I adapted the cifar10...

The first number is your image number in the batch. Say you fed two images, you may find 2 objects in image 1 and 1 object in image 2. (1,20,454)...

OK, so I fixed one of the issues: I installed keras and keras2onnx, and ocnverted both models with opset =10 Now both models function exactly the same: The batch parameter...

The thing is in your old repo (yolov5) __getitam_ has all the code to and I can trace everything and I've adapted it to read read my data as it...

ok found the conversion in ops.py and made a div 0 error to have it trace it: It uses cv2.minAreaRect(pts) . btw the function comment is wrong saying rotation has...

Nope, because for some obscure reason PTS is not the 4 xy points given in the label file, but magically it got turned into 100 points! The doc are wrong....

I just trained with resample = 5 , and disabled patience to allow training for 2000 epochs more after the metrics stop improving, but the box loss keeps improving. Also...