Hi
Detectnet-camera facenet of jetson-inference is being tested.
I want to detect the face by putting the Res10_300x300_SSD_iter_140000.caffemodel that has been re-learned by ResnetSSD instead of the corresponding facenet-120 network model snapshot_iter_24000.caffemodel.
However, as a result of executing NvCaffe parser, there was an error “could not parse layer type Normalize”.
Please let me know whether I approached it in the wrong way.
Please let me know if it is possible to do only caffe model learned by detectNet
Hi,
jetson_inference will run your model with TensorRT library.
So if a model is fully supported by the TensorRT, it can be executed with jetson_inference.
Unfortunately, SSD is not a fully supported model and you will need to integrate several plugin to enable it.
We also have a sample to demonstrate how to integrate a custom layer into TensorRT.
Please check this sample for more information: /usr/src/tensorrt/samples/sampleSSD/
Thanks.
Hi um,
I wasted two weeks testing, only Nvidia provided or dedicated models works :) Most others either don’t work or of very poor performance。
I stopped playing with nano, get back working on opencv and other meaningful tasks.
Hi,
Sorry to hear this.
We keep implementing new operations but still some non-covered variance.
We already have a sample for the SSD model. It’s recommended to give it a try.
Sorry for any inconvenience it brings to you.
Thanks.
Hi,
Sorry that I saw your post on another topic.
If you are finding a python SSD example, please check this one:
[url]Sample Support Guide :: NVIDIA Deep Learning TensorRT Documentation
Thanks.
Hi ktktkkt
I am also like you.
But NVIDIA thinks it will bring us something more advanced and improved performance.
Hi AastaLLL
Your answer helped me.
Thanks you.