To detect different crops a large set of photos need to be taken and boundary boxes 'drawn' around the actual plant to help determine where it is in the camera frame. Since we dont actually have any newly planted crops at this time of year, I've used a ready prepared set of dog photos as a practice run. These are accurate step by step instructions and this text assumes all the relevant software is already installed on the Jetson:

Prerequisites: 

Jetson TX2 flashed with JetPack 3.3.

Caffe version: 0.15.14

DIGITS version: 6.1.1

Check that all software is installed correctly by using the pre-installed dog detect model that comes with Jetpack by running this in terminal:

$ sudo ~/jetson_clocks.sh && cd jetson-inference/build/aarch64/bin && ./detectnet-camera coco-dog

It will take a few minutes to load up before the camera footage appears.

To start from scratch with a set of photos, first turn on the DIGITS server:

$ sudo ~/jetson_clocks.sh && cd digits && export CAFFE_ROOT=/home/nvidia/caffe && ./digits-devserver

Now we're going to build the model using actual images of dogs with their associated text files:

In browser naviate to http://localhost:5000/      

Importing the Detection Dataset into DIGITS: 

> Datasets > Images > Object Detection

Training image folder:  /media/nvidia/2037-F6FA/coco/train/images/dog 

Training label folder:  /media/nvidia/2037-F6FA/coco/train/labels/dog 

Validation image folder: /media/nvidia/2037-F6FA/coco/val/images/dog 

Validation label folder: /media/nvidia/2037-F6FA/coco/val/labels/dog 

Pad image (Width x Height): 640 x 640 Custom classes: dontcare, dog 

Group Name: MS-COCO Dataset Name: coco-dog

> Create > Home > Models > Images > Object Detection

> Select Dataset: coco-dog 

Training epochs = 16
Snapshot interval (in epochs) = 16
Validation interval (in epochs) = 16

Subtract Mean: none 

Solver Type: Adam 

Base learning rate: 2.5e-05 

> Show advanced learning options 

Policy: Exponential Decay 

Gamma: 0.99 

batch size = 2 

batch accumulation = 5  (for training on Jetson TX2)

Specifying the DetectNet Prototxt: 

> Custom Network > Caffe 

The DetectNet prototxt is located at /home/nvidia/jetson-inference/data/networks/detectnet.prototxt in the repo.

> Pretrained Model = /home/nvidia/jetson-inference/data/networks/bvlc_googlenet.caffemodel

 >Create 

Location of epoch snapshots: /home/nvidia/digits/digits/jobs You should see the model being created through a series of epochs. Make a note of the final epoch.

Navigate to /home/nvidia/digits/digits/jobs and open the latest job folder and check it has the 'snapshot_iter_*****.caffemodel' files in it. Make a note of the highest '*****' number then copy and paste the folder into here for deployment: /home/nvidia/jetson-inference/build/aarch64/bin.

Rename the folder to reflect the number of epochs that it passed, eg myDogModel_epoch_30.

For Jetson TX2, at the end of deploy.prototxt, delete the layer named cluster:

layer {
  name: "cluster"
  type: "Python"
  bottom: "coverage"
  bottom: "bboxes"
  top: "bbox-list"
  python_param {
    module: "caffe.layers.detectnet.clustering"
    layer: "ClusterDetections"
    param_str: "640, 640, 16, 0.6, 2, 0.02, 22, 1"
  }
}

Open terminal and run, changing the '*****' number accordingly:

$ cd jetson-inference/build/aarch64/bin && NET=myDogModel_epoch_30 && ./detectnet-camera \
--prototxt=$NET/deploy.prototxt \
--model=$NET/snapshot_iter_*****.caffemodel \

Hit return twice and you'll see various messges including: 

[TRT]  attempting to open cache file dogPoo_epoch_8/snapshot_iter_3088.caffemodel.2.tensorcache 

[TRT]  cache file not found, profiling network model

This is not an error!

If you've got: 

[TRT]  building CUDA engine 

Then all is good - just wait a few minutes for it to complete and then the camera should activate.

Now find / borrow a dog and test for bounding boxes!