Detecting multiple object classes simultaneously using Driveworks

adjyt · March 21, 2017, 4:55am

Hello All,

I have gone through the “object detection” sample provided along with the Driveworks SDK. It helps to detect “cars” in an image and creates the bounding boxes. I am wondering if there is any way via which I can detect multiple object classes (with bounding boxes) in a single picture using Driveworks.
I understand that I need to change the caffe model for this task. Will it be possible to use DetectNet model for such kind of activity.
Please suggest.

Thanks
Deepika

SteveNV · March 22, 2017, 1:11am

Hello adjyt,

“object detetion” sample only support single object like car.
If you want to use multi class object detection, please use “drivenet” sample instead of object detection. Thanks.

nvok · March 27, 2017, 2:14pm

Hi Deepika,

Take a look at this thread here: Car-Pedestrian multi class object detection DetectNet mAP and inference · Issue #1359 · NVIDIA/DIGITS · GitHub
It contains a link to the prototxt of a two-class version of DetectNet.

bpinaya · March 30, 2017, 2:31am

hey adjyt, I had good results with this:

github.com

NVIDIA/caffe/blob/caffe-0.15/examples/kitti/detectnet_network-2classes.prototxt

# DetectNet network

# Data/Input layers
name: "DetectNet"
layer {
  name: "train_data"
  type: "Data"
  top: "data"
  data_param {
    backend: LMDB
    source: "examples/kitti/kitti_train_images.lmdb"
    batch_size: 10
  }
  include: { phase: TRAIN }
}
layer {
  name: "train_label"
  type: "Data"
  top: "label"
  data_param {

This file has been truncated. show original

You just need to retrain the network and you are all set, good luck!

adjyt · March 30, 2017, 6:58am

hi nvok/bpinaya,

I used the “detectnet_network-2classes.prototxt” file to train my network on Kitti data.
But I am getting the following error while training:
<<
error code -11
Train net output #0: loss_bbox = 1.12773 (* 2 = 2.25547 loss)
Train net output #1: loss_coverage = 4.84529 (* 1 = 4.84529 loss)
Iteration 1264, lr = 0.0001
Snapshotting to binary proto file snapshot_iter_1276.caffemodel
Snapshotting solver state to binary proto file snapshot_iter_1276.solverstate
Iteration 1276, Testing net (#0)
Ignoring source layer train_data
Ignoring source layer train_label
Ignoring source layer train_transform

Any suggestions?

Thanks
Deepika

nvok · March 30, 2017, 3:19pm

Hi Deepika,

Please move your question to a DIGITS thread, at best to the one that I posted earlier (Car-Pedestrian multi class object detection DetectNet mAP and inference · Issue #1359 · NVIDIA/DIGITS · GitHub). You will receive the best help from there.

Thanks,
Oliver

a.a.menezes · March 1, 2020, 1:42pm

Dear Steve,
Can Drivenet Sample import and use Detectnet? I think this sample is only suitable for using Drivenet proprietary model.

Secondly can dwObjectDetector, dwObjectClustering functions be used with Detectnet? I think these are solely for Drivenet since they require a Drivenet Handle.

shayNV · March 1, 2020, 7:02pm

Hi a.a.menezes,

Unfortunately, The DriveNet Sample and its interface are not changeable, as the binaries and libraries are precompiled and the network is built in the software stack.

dwObjectDetector, dwObjectClustering functions can be used only with DetectNet network, as you said because the initialization function expects the DriveNet Handle(dwDriveNetHandle_t) which has its own specific internal structure.