Problem with detection

[emasterclassacademy]

Problem with detection

[Intruder31]

Same issue. I have a custom trained YOLOv3 model, converted that to tf format using the convert.py in this repo; doesn't detect anything. Are you using custom trained model too?

For me the pre-trained YOLOv3 works though.

[PiotrG1996]

I used official weights from this source: https://pjreddie.com/darknet/yolo/

None of them worked for me. Could you share .tf files? I have no idea what's wrong with my files.

[Intruder31]

Sure I could share them. What platform do you want me to upload them on?

Also, would love to take a look at your .tf files to resolve my issue with custom trained weights.

I have attached the list of files as well as their sizes (if that helps) after conversion. The highlighted ones are newly formed.

[Intruder31]

Ok I guess I'd downloaded the wrong weights and I'm guessing you have too.

You will need to download weights from https://pjreddie.com/media/files/yolov3.weights (~248 MB) and then convert them to .tf

To train on custom datasets, we need convolutional weights that are pre-trained on Imagenet. These are found at https://pjreddie.com/media/files/darknet53.conv.74 (~162.5 MB)

[PiotrG1996]

Thanks, this fixed my issue. Nonetheless, this algorithm seems to be very memory consuming. GPU results on gtx 1080 2-3 fps.
Have you tested tensorflow lite with it?