- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I have download openvino_toolkit_runtime_raspbian_p_2021.4.752 on raspberry pi4b. I want to perform yolov3-tiny on raspberry pi4B with NCS2.But when i try it on Ubuntu18 ,some mistakes happend.
The demo i use is ./intel64/Release/object_detection_demo
I get yolov3-tiny.weights in pjreddie/darknet: Convolutional Neural Networks (github.com)
l really need help
thanks
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi fizzgo,
Have you tried with OpenVINO™ Toolkit Public Pre-Trained Model, yolo-v3-tiny-tf ?
Besides, please also share the Model Optimizer command in converting your yolo-v3 model.
Regards,
Peh
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Second,l tried yolov3-tiny (I got it from pjreddie).I also converted it successfully,but it ran wrongly.I have described it in my first question.
At last,l tried my own yolov3-tiny model(only one class),it detected nothing.The following pictures shows my orders .
Thanks for your help.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
First,l have tried yolo-v3-tiny-tf ,but it performed badly.
I got two files yolo-v3-tiny-tf.json and yolo-v3-tiny-tf by the order : python3 downloader.py --name yolo-v3-tiny-tf
then i used mo_tf.py to convert it to IR(xml bin) successfully.But when i tried it with NCS2,it act like below:
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi fizzgo,
For the OpenVINO™ Toolkit Public Pre-Trained Model, yolo-v3-tiny-tf, you are getting poor inferencing performance due to the model conversion.
You can use the converter.py script to convert the model or manually use mo_tf.py script to convert with full Model Optimizer arguments:
python3 mo_tf.py --input_shape=[1,416,416,3] --input=image_input --scale_values=image_input[255] --reverse_input_channels --transformations_config=$dl_dir/yolo-v3-tiny-tf/yolo-v3-tiny-tf.json --input_model=$dl_dir/yolo-v3-tiny-tf/yolo-v3-tiny-tf.pb
Next, convert the yolo-v3-tiny model ( get from Pjreddie) with the same Model Optimizer arguments especially the --reverse_input_channels argument.
Lastly, for the custom yolo-v3-tiny model (one class), convert the model with the same Model Optimizer arguments and make sure you have modified the classes in yolo_v3_tiny.json.
Regards,
Peh
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Dear Peh,
Thanks for your help,i have solved my problems.
Regards,
fizzgo
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi fizzgo,
This thread will no longer be monitored since this issue has been resolved. If you need any additional information from Intel, please submit a new question.
Regards,
Peh

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page