Does Deepstream 5.1 support YOLOV4 .etlt file exported from Nvidia’s Transfer Learning Toolkit?

Similar question to Does Deepstream 5.0 support YOLOV4 .etlt file exported from Nvidia's Transfer Learning Toolkit?

Is there any update?

@abrar.shahriar

The answer is No for both DS 5.0 and 5.1, you must use tlt-converter to build the .engine file from the .etlt file, quoted from the doc:

To integrate a model trained by TLT with DeepStream, you shoud generate a device-specific optimized TensorRT engine using tlt-converter . The generated TensorRT engine file can then be ingested by DeepStream (Currently, YOLOv4 etlt files are not supported by DeepStream).

After conversion to .engine will the default object parser work “out of the box” or do we need to use some sort of custom parser. Is this parser(GitHub - NVIDIA-AI-IOT/yolov4_deepstream) compatible?

You still need the TRT OSS and the custom parser from this repo to run the .engine file produced by tlt-converter. The info is buried in this part of the doc.

1 Like

YoloV4 is in TLT3.0 which needs to work with DS5.1.

And, as mentioned in GitHub - NVIDIA-AI-IOT/deepstream_tlt_apps at release/tlt3.0 , TRT OSS is needed for YOLOV3 , YOLOV4 if you are using TRT version(such as TRT7.1) before TRT7.2 .

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.