I was able to convert the .pb file of YOLO object detection API to IR files(.xml and .bin). But when I try to run inference using object_detection samples I am getting the following errors:
./object_detection_sample_ssd -d "CPU" -m /home/gw2/IRfiles/tiny-yolo-voc-1c.xml -i /home/gw2/IRfiles/ATD_test_5.jpg
[ ERROR ] Can't find a DetectionOutput layer in the topology
./object_detection_demo_ssd_async -d "CPU" -m /home/gw2/IRfiles/tiny-yolo-voc-1c.xml -i /home/gw2/final.mp4
[ ERROR ] No such parameter name 'num_classes' for layer output/YoloRegion
3)Even object_detection_sample will not work
Can someone guide on performing inference for YOLO?
Divyashree, first, how did you get the YOLO *.pb file ? Was it based on yolov3 or yolov2 ? Did you use deployment_tools\documentation\docs\YOLOTF.html as a guide ? The reason I am asking is because there is a known (not related to Intel) bug that prevents one from converting a a Yolov3 to a tensor flow graph. More details about this bug here. I faced this bug myself when I used darkflow as YOLOTF.html describes.
Thank you for using OpenVino and I shall await your reply.
Hi Shubha, Thanks for your reply..
I used Tiny Yolo and then I converted the corresponding cfg and weights to .pb.
Yes, I used YOLOTF.html as a reference to generate IR files.
Please, help me with the inference part.