I trained a custom object detection with SSD and MobilenetV2 starting from Tensorflow model zoo (http://download.tensorflow.org/models/object_detection/ssd_mobilenet_v2_coco_2018_03_29.tar.gz) and I converted it with OpenVino. If I run the model with python example, on Neural Computer stick it works correctly, but if I run it with c++ example "object_detection_sample_ssd.exe", the inference is done correctly, but the results of detections have non sense. Some suggestion about?
I solved by myself. The demo is tuned for Cafe model, that use BRG images, converting the image in RGB the detection being ok.
Should be helpful to add an option to choose the input color format depending on model