Intel® Distribution of OpenVINO™ Toolkit
Community support and discussions about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all things computer vision-related on Intel® platforms.

Converting a .pb file to NCS graph

Community Manager



I was using Darkflow to detect multiple objects from a video. I used a model trained on the COCO dataset (yolo.coco.cfg and yolo.coco.weights).


I successfully converted this model to a protobuf .pb file to be used with Tensorflow using:


flow --model cfg/yolo.coco.cfg --load bin/yolo.coco.weights --savepb


I got two files yolo.coco.pb and yolo.coco.meta


And this is working fine and I was able to load this model in to Tensorflow and run detections.


Now, I am trying to convert this .pb file to a frozen graph to be used with NCS. But looks like it needs a checkpoint file.


Since I am using pre-trained weights, I don't have the checkpoint file with me.


How can I convert this .pb file to be used with NCS?


Is it necessary that I train the model?


I'm a bit new to this, so please forgive any mistakes.



0 Kudos
2 Replies
Community Manager

Same issue here. Have you got any clue about how to solve this problem?

Community Manager

Hi @rahules and @zdx66536,


Thanks for reaching out. May I ask which NCSDK version are you guys using? I've seen people using YOLO on the NCS using Darkflow, I was able to get the graph file to work on the NCS, find the relevant URLs below:





If you need help troubleshooting please provide more details on your setup (host OS, VM?, NCSDK version, etc.), the list of steps you are following, and/or any other information that could help us. We will be waiting for your reply.


Best Regards,