I trained a model with tiny-yolo-voc.weights to detect number plates. The training was done using Darkflow. The detection work's fine with .pb and .meta files, for which i used return_predict function(https://github.com/thtrieu/darkflow/blob/b2aee0000cd2a956b9f1de6dbfef94d53158b7d8/darkflow/net/flow.py#L76) . Then I compiled, the .pb file to generate the graph file(My ncsdk is installed in an aws ec2 instance). After obtaining Inference using the graph.queue_inference_with_fifo_elem( ) method , I referred this work ( https://github.com/fernandodelacalle/yolo-darkflow-movidius/blob/master/detection_camera.py, https://github.com/fernandodelacalle/yolo-darkflow-movidius/blob/8a2c07878e98fe370c72f8a6e951041fbb5093c0/yolo_utils.py ) for generating bbox from graph's output. Which is in fact an adaptation of return_predict method in darkflow. But no boxes were detected. Can anyone point out why this is happening?