OpenVINO model zoo models compatibility with Neural Compute Stick 2

[naman]

Can we use models like mask_rcnn_resnet50_atrous_coco, yolact-resnet50-fpn-pytorch and yolo-v4-tf from OpenVINO modelzoo to perform inference on Intel® NCS 2? If so what FPS can we get? Does Intel® NCS 2 support the ncappzoo apps only?

[naman]

D

[Wan_Intel]

Hi Naman Mehta,

Thank you for reaching out to us.

You can use yolact-resnet50-fpn-pytorch and yolo-v4-tf models for inference using Intel® Neural Compute Stick 2.

I’ve tested these models with Intel® Neural Compute Stick 2 and I share here the FPS values which were obtained using Benchmark C++ Tool.

Model	Throughput
yolact-resnet50-fpn-pytorch	1.88
yolo-v4-tf	1.40

For mask_rcnn_resnet50_atrous_coco model, when I tested using the input shape arguments provided in model.yml, I observed the system runs out of memory. However, when I change the input shape to [1,3,224,224], I obtained 0.16 FPS. 

Performance benchmarks of selected models across different Intel hardware is available in Intel® Distribution of OpenVINO™ toolkit Benchmark Results.

Regards,

Wan

[Wan_Intel]

Hi Naman,

This thread will no longer be monitored since we have provided a solution. 

If you need any additional information from Intel, please submit a new question.

Regards,

Wan