Wrong output of YOLO v2-tiny using OpenVINO

[Dandriyal__Prashant]

Hi, 

I am using the YOLO v2-tiny model (pre-trained on COCO dataset) and have obtained the IR files using the command

```

pradan@pradan-HP-15-Notebook-PC:~/Desktop/intel/openvino_2020.1.023/deployment_tools/model_optimizer$ python3 mo.py --input_model ~/Converted-YOLO-Models/yolo-v2-tiny-coco.pb --batch 1 --scale 255 --output_dir ~/Desktop/intel/openvino_2020.1.023/Intel-EdgeAI-Nanodegree/PeopleCounterApp/popo_models --reverse_input_channels --tensorflow_use_custom_operations_config ~/Desktop/intel/openvino_2020.1.023/deployment_tools/model_optimizer/extensions/front/tf/yolo_v2_tiny.json

```

I am also able to make the inference and obtain the output blob. I am learning about the anchors and things related to the conversion of output to bounding boxes, but I feel that I am getting the pre-processing wrong... or the predictions are just NOT right; I have shared the results below. 

I feel that there is something wrong in the RGB->BGR conversion or some other pre-processing part. Please help 

[JAIVIN_J_Intel]

Hi Prashant,

Have you tried without using the --reverse_input_channels parameter?

Please refer the documentation about When to Reverse Input Channels.

Regards,

Jaivin

[Dandriyal__Prashant]

Hey @Jalvin James, I doubt that the YOLOv3 conversion doesn't require that. So, I didn't try it for v2-tiny also. 

[JAIVIN_J_Intel]

Since you have used --reverse_input_channels parameter on the command mentioned, please try removing the parameter and then convert the model again.

Also, please provide information about the following:

• Are you using yolo-v2-tiny from the open model zoo?
• Are you running the model on any Openvino Samples?

Regards,

Jaivin

[Dandriyal__Prashant]

No. I used the official weights hosted at PjReddie's site.
I don't understand how I could run it on the samples. Please explain:)