Refer to https://github.com/NVIDIA/TensorRT/tree/master/samples/opensource/sampleUffSSD

### Step 1 ###
### Download TensorFlow SSD-Inception_v2_coco model 

$ cd ~/Downloads
$ wget http://download.tensorflow.org/models/object_detection/ssd_inception_v2_coco_2017_11_17.tar.gz
$ tar -zxvf ssd_inception_v2_coco_2017_11_17.tar.gz

### Step 2-1 ###
### Copy the TensorFlow protobuf file (frozen_inference_graph.pb) from the downloaded directory in the previous step 
### to the working directory (for example /usr/src/tensorrt/samples/sampleUffSSD/).

### [sudo] password for adlink: adlink
$ cd ~/Downloads
$ echo adlink | sudo -S cp ssd_inception_v2_coco_2017_11_17/frozen_inference_graph.pb /usr/src/tensorrt/samples/sampleUffSSD/frozen_inference_graph.pb
$ cd /usr/src/tensorrt/samples/sampleUffSSD

### Step 2-2 ###
### Perform preprocessing on the tensorflow model using the UFF converter. 

$ echo adlink | sudo -S python3 /usr/lib/python3.6/dist-packages/uff/bin/convert_to_uff.py frozen_inference_graph.pb -O NMS -p config.py

### Step 3 ###
### Run the following command for the conversion
### The conversion takes about 3 minutes.

$ /usr/src/tensorrt/bin/trtexec --uff=frozen_inference_graph.uff --output=NMS --uffInput=Input,3,300,300 --workspace=3000 --saveEngine=/home/adlink/Downloads/model/ssdv2.engine --verbose --maxBatch=4

### Step 4 ###
### Test by EVA

$ cd ~/Downloads
$ gst-launch-1.0 pylonsrc camera=0 fps=15 ! videoscale ! video/x-raw, width=800, height=600 ! videoconvert ! adrt model=/home/adlink/Downloads/model/ssdv2.engine batch=1 device=0 scale=0.0078 mean="0 0 0" norm=false ! adtrans_ssd label=/home/adlink/Downloads/model/ssd_coco_labels.txt ! admetadrawer ! videoconvert ! fpsdisplaysink video-sink=xvimagesink text-overlay=true