how to test perception models independently

[fenkywuk]

I want to test the perception models independently in Apollo, for example, by inputting some images to the perception models and see what I get. I have successfully tested the traffic light recognition model (which has a simple and common network architecture) with a normal Caffe. (I use pycaffe for this testing)

However, I found that models like traffic light detection / object detection used in Apollo includes some customized layers, and the used Caffe is not open-sourced. In this case, how can I test these models independently?

Thanks!

[liouyx]

Source code of historical models are not opened. The new models will be released in September, and the training source code, evaluation dataset and deployment method will be opened. You can test new models independently then.