Using DetectNet caffe model in sample_object_detector

Hi,

(Reposting my question from https://devtalk.nvidia.com/default/topic/998570/jetson-tx1/could-not-parse-layer-type-python/post/5108818/#5108818 as requested by AastaLLL)

I’m trying to train my own DetectNet caffe model and run it using the sample_object_detector software provided in the DriveWorks SDK. I’ve trained my caffe model using the default DetectNet network (https://github.com/NVIDIA/caffe/blob/caffe-0.15/examples/kitti/detectnet_network.prototxt), but commenting out the last 4 python layers (cluster, cluster_gt, score, mAP) because both TensorRT and nvCaffe do not appear to support python layers.

However, after using TensorRT to optimize the model, running the TensorRT binary with sample_object_detector does not return any bounding boxes, whereas using the default TensorRT binary provided in sample_object_detector does.

How should we modify the DetectNet network / sample_object_detector source file to get the object detector to work with our own network?

Thanks!

Hello ruijie,

Could you please refer to below link “How to use DRIVE PX 2” part(page 114 ~)? Thanks.
https://developer.nvidia.com/driveworks/getting-started-with-drivepx2

Hi mania91,

Those instructions are very helpful, and enabled me to train and use my own network.

Thanks!

Hello SteveNV,

could you update the link you mentioned, please? As I, too, would like to use a network, trained on my own.

Would be great!
Thanks!