Model acuracy in the Jetson reference

Autonomous Machines Jetson & Embedded Systems Jetson TX2
1

I’m practicing “Re-training on the Cat/Dog Dataset” at the jetson-inference. One of them is a model acuracy. How can I print out that graph?

2

And I want to retrain ssd-mobilenev2. In the example of using pytorch, there is only one about classification.

And the --arch option also has a mobileenet_v2 but no ssd-mobilenet_v2.

Then, go to jetson-inference/python/training/detection and try to get an error when you go to

python3 train.py --model-dir=cat_dog ~/datasets/cat_dog

.

All the codes shown in the example are well executed, but if you go to detection, it won’t work.
The error is unrecognited argumets : /home/hybin/datastes/cat_dog.

How should i solve these problems?

3

Hi,

How do you train a customized model.
If you are using DIGITs, you can find the similar accuracy graph on the interface.

For the pyToch problem, would you mind to check this issue first?
https://github.com/dusty-nv/jetson-inference/issues/370

Thanks.

4

I train at “Transfer Learning with PyTorch” in “Hello AI world.” There is no problem with the pytorch.

I would like to use ssd-mobilenetv2 of detection rather than resnet-18 of the classification shown in the example.
The resnet-18 of the classification works perfectly.

5

Hi,

Do you follow this page to retrain a resnet-18 model?
https://github.com/dusty-nv/jetson-inference/blob/master/docs/pytorch-cat-dog.md

Please noticed that the model used in this page is classifier.
And ssd-mobilenetV2 is a detector, which is not compatible in this tutorial.

Thanks.