### Step 1 ###
### Download googlenet caffe model 

$ cd /home/adlink/Downloads/model
$ wget http://dl.caffe.berkeleyvision.org/bvlc_googlenet.caffemodel
$ wget https://raw.githubusercontent.com/BVLC/caffe/master/models/bvlc_googlenet/deploy.prototxt

### Step 2 ###
### Caffemodel contain two files : prototxt file and caffemodel file. 
### To inference we normally require the deploy.prototxt file.
### When convert Caffe model, we need to input the output layer name. 
### In the above link, the input layer is "Input" and the output layer is "prob"
### The conversion takes about 1 minutes.

$ cd /home/adlink/Downloads
$ /usr/src/tensorrt/bin/trtexec --model=/home/adlink/Downloads/model/bvlc_googlenet.caffemodel --deploy=/home/adlink/Downloads/model/deploy.prototxt --output=prob --workspace=3000 --maxBatch=4 --verbose --saveEngine=/home/adlink/Downloads/model/googlenet.engine

### Step 3 ###
### Test by EVA

$ gst-launch-1.0 pylonsrc camera=0 fps=20 ! videoconvert ! adrt model=/home/adlink/Downloads/model/googlenet.engine batch=1 ! adtrans_classifier label=/home/adlink/Desktop/EVA_IDE/model/googlenet-v2_RT_labels.txt ! admetadrawer ! videoconvert ! fpsdisplaysink video-sink=xvimagesink text-overlay=true