How can I evaluate the mAP of a frozen graph using the tensorflow object detection API?

I have tried loading the frozen graph instead of the checkpoints editing the eval*.py but seems like it requires variables like "global_step" etc.