TDA4VM: some question about tda4

user6065254

Part Number: TDA4VM
Other Parts Discussed in Thread: TDA2

Hi:

I was using TDA2 platform to develop object detection(ssd-jacinto) application,

but by my testing (link), the model latency run on TDA2 platform will be above 130 ms (27-fps) which not satisfied my requirement (less than 1000/30 ms).

However, my boss ask me evaluating the TDA4 platform, so a few question below:

1. Did anyone test/verify ssd-jacinto model on TDA4 platform?

2. How about ssd-jacinto model latency time running on TDA4 platform?

3. Is TDA4 platform capable of running ssd-jacinto model above 30-fps?

4. What is SDK supporting the TDA4 platform?

Thanks

Best / Regards

over 5 years ago

0 Manu Mathew over 5 years ago

TI__Genius 11226 points

You can find a lot of TDA4 resources here: https://www.ti.com/product/TDA4VM

(Including specifications, application notes and information about software).

Specifically, information about SDK is here: https://www.ti.com/tool/PROCESSOR-SDK-DRA8X-TDA4X

Information about training Deep Learning Models (including Semantic Segmentation, Object Detection etc) is here: https://github.com/TexasInstruments/jacinto-ai-devkit

We provide training examples and scripts for training SSD and RetinaNet object detectors using the repository pytorch-mmdetection. We also have scripts to export the model into a format (.onnx and .prototxt) that TIDL can understand.

I hope this information helps you to get started.

Processors

Processors forum

TDA4VM: some question about tda4