# [09/12] Jetson Nano

## 10:30 - 11:50
### Jetson Nano 安裝與測試
#### [Jetson-stats](https://github.com/rbonghi/jetson_stats)

$ sudo apt install python-pip python3-pip
$ sudo pip3 install jetson-stats

###### Reboot 重開機

# Open the jtop prompt interface
$ jtop

# Show the status and all information about your NVIDIA Jetson NANO
$ jetson_release

# Another command for showing the status of Jetson NANO
$ export | grep JETSON

# Ubuntu version
$ lsb_release -a

#### CUDA

# Copy cuda samples to root directory
cp -a /usr/local/cuda-10.2/samples/ ~/

# Test CUDA using deviceQuery test tool
cd ~/samples/1_Utilities/deviceQuery
make
./deviceQuery

# Run CUDA Sample, OceanFFT
cd ~/samples/5_Simulations/oceanFFT
make
./oceanFTT

# Run CUDA Sample, smokeParticles (take about ~1 min to compile it.)
cd ~/samples/5_Simulations/smokeParticles/
make
./smokeParticles

# Run n-body simulation
cd ~/samples/5_Simulations/nbody/
make
./nbody

# You can follow the example steps above, and run other applications

#### VisionWorks

# For more practical usage of GPU, here are some demostrations of visionworks. 
# Ref: https://developer.nvidia.com/embedded/visionworks

# Copy samples
mkdir ~/visionworks
/usr/share/visionworks/sources/install-samples.sh ~/visionworks/

# Hough detecion
cd ~/visionworks/VisionWorks-1.6-Samples/demos/hough_transform/
make
~/visionworks/VisionWorks-1.6-Samples/bin/aarch64/linux/release/nvx_demo_hough_transform

# Motion estimation
cd ~/visionworks/VisionWorks-1.6-Samples/demos/motion_estimation/
make
~/visionworks/VisionWorks-1.6-Samples/bin/aarch64/linux/release/nvx_demo_motion_estimation

# You can follow the example steps above, and run other applications

---
## 15:20 - 16:00
### 遷移學習Transfer learning --- Classification
#### 0. Enviroment

$ git clone --recursive https://github.com/dusty-nv/jetson-inference
$ cd jetson-inference
$ docker/run.sh

# exit docker 
$ exit

#### 1. Download the Cat/Dog dataset

$ cd jetson-inference/python/training/classification/data

# Download the dataset
$ wget https://nvidia.box.com/shared/static/o577zd8yp3lmxf5zhm38svrbrv45am3y.gz -O cat_dog.tar.gz

# Untar the dataset
$ tar xvzf cat_dog.tar.gz

#### 2. Train the model

$ cd jetson-inference/python/training/classification
$ python3 train.py --model-dir=models/cat_dog data/cat_dog

>* Note : 停止訓練可按 Ctrl+C

#### 3. Load trained 100 epoch model
[**Download Link**](https://nvidia.app.box.com/s/zlvb4y43djygotpjn6azjhwu0r3j0yxc)

# In ~/Downloads/

# Copy the model
$ cp cat_dog_epoch_100.tar.gz ~/jetson-inference/python/training/classification/models/cat_dog

# Untar the model
$ cd /jetson-inference/python/training/classification/models/cat_dog
$ tar xvzf cat_dog_epoch_100.tar.gz

# Evaluate the model
$ cd /jetson-inference/python/training/classification
$ python3 train.py --resume=models/cat_dog/cat_dog_epoch_100/model_best.pth.tar data/cat_dog --evaluate 

#### 4. Converting the Model to ONNX
[Open Neural Network eXchange](https://onnx.ai/) (ONNX) is an open standard format for representing machine learning models.

$ python3 onnx_export.py --model-dir=models/cat_dog/cat_dog_epoch_100

This will create a model called ```resnet18.onnx``` under ```jetson-inference/python/training/classification/model/cat_dog/cat_dog_epoch_100/```

#### 5. Processing Images with TensorRT
[TensorRT](https://developer.nvidia.com/tensorrt) is an SDK for high-performance deep learning inference

$ imagenet.py --model=models/cat_dog/cat_dog_epoch_100/resnet18.onnx --input_blob=input_0 --output_blob=output_0 --labels=data/cat_dog/labels.txt /jetson-inference/data/images/cat_1.jpg /jetson-inference/data/images/test/cat_1.jpg

#### 6. Running the Live Camera with TensorRT

$ imagenet.py --model=models/cat_dog/cat_dog_epoch_100/resnet18.onnx --input_blob=input_0 --output_blob=output_0 --labels=data/cat_dog/labels.txt /dev/video0

#### 7. Download the models

$ cd /jetson-inference/tools
$ ./download-models.sh

# Your models is stored in /jetson-inference/data/network

---