Intel® Distribution of OpenVINO™ Toolkit
Community support and discussions about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all things computer vision-related on Intel® platforms.
6059 Discussions

DL Streamer: Unable to get detections from Yolov3 model

Shantanu_Bakare
Beginner
358 Views

Hi,

I am working with DL streamer pipeline on openvino version '2021.3.394'.

When running the given python sample with the default detection model, i.e. face-detection-adas-0001,  I am able to get a tensor layer as 'detection_out'.

But in the same sample code, when I change this to 'yolo-v3-tf.xml', there are no tensor layers available. I also tried checking for different layer names for yolo from here:  yolo-v3 

I also looked into yolo-v3-tf.xml file and searched for potential layer names, but couldn't find a working one.

How can I get detections from this sample code using yolo-v3-tf.xml model?

Labels (1)
0 Kudos
1 Solution
Zulkifli_Intel
Moderator
329 Views

Hello Shantanu_Bakare,

Thank you for reaching out to us.

 

The Face Detection And Classification Python Sample uses the following pre-trained models from OpenVINO Open Model Zoo.

·      face-detection-adas-0001

·      age-gender-recognition-retail-0013

·      emotions-recognition-retail-0003

·      facial-landmarks-35-adas-0002-0009

·      head-pose-estimation-adas-0001

 

This sample is used to detect faces and it is not suitable to use with the yolo-v3-tf model as the model was trained to detect objects.

You can run the yolo-v3-tf model with dlstreamer directly using this command:

 

gst-launch-1.0 \

filesrc location=${VIDEO_EXAMPLE} ! decodebin ! \

gvadetect model=${DETECTION_MODEL} model_proc=${DETECTION_MODEL_PROC} device=CPU ! queue ! \

gvawatermark ! videoconvert ! fpsdisplaysink video-sink=xvimagesink sync=false

 

Replace VIDEO_EXAMPLE with your input sample, DETECTION_MODEL with the path to yolo-v3-tf model, and DETECTION_MODEL_PROC with /opt/intel/openvino_2021.3.394/data_processing/dl_streamer/samples/model_proc/yolo-v3-tf.json

 

I tested with a bottle-detection.mp4 sample video. This is my result:

dlstreamer.PNG

 

View solution in original post

2 Replies
Zulkifli_Intel
Moderator
330 Views

Hello Shantanu_Bakare,

Thank you for reaching out to us.

 

The Face Detection And Classification Python Sample uses the following pre-trained models from OpenVINO Open Model Zoo.

·      face-detection-adas-0001

·      age-gender-recognition-retail-0013

·      emotions-recognition-retail-0003

·      facial-landmarks-35-adas-0002-0009

·      head-pose-estimation-adas-0001

 

This sample is used to detect faces and it is not suitable to use with the yolo-v3-tf model as the model was trained to detect objects.

You can run the yolo-v3-tf model with dlstreamer directly using this command:

 

gst-launch-1.0 \

filesrc location=${VIDEO_EXAMPLE} ! decodebin ! \

gvadetect model=${DETECTION_MODEL} model_proc=${DETECTION_MODEL_PROC} device=CPU ! queue ! \

gvawatermark ! videoconvert ! fpsdisplaysink video-sink=xvimagesink sync=false

 

Replace VIDEO_EXAMPLE with your input sample, DETECTION_MODEL with the path to yolo-v3-tf model, and DETECTION_MODEL_PROC with /opt/intel/openvino_2021.3.394/data_processing/dl_streamer/samples/model_proc/yolo-v3-tf.json

 

I tested with a bottle-detection.mp4 sample video. This is my result:

dlstreamer.PNG

 

Zulkifli_Intel
Moderator
306 Views

Hello Shantanu Bakare.


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.


Sincerely,

Zulkifli


Reply