jetson nanoで物体検知を行う
jetson nanoで物体検知を試した
日本語の情報がどのあたりが最新か不明のためでいろいろ手間取った ssh経由でやVNCでやろうとしたけど結局直接端末でやらないと面倒そうなので端末で制御
パッケージインストール
sudo apt-get update sudo apt-get install git cmake libpython3-dev python3-numpy
jetson-interfaceを利用して画像認識を行うため、git clone
git clone --recursive https://github.com/dusty-nv/jetson-inference cd jetson-inference mkdir build cd build cmake ../
↓
モデルダウンロード
# modelを選択する（デフォルトのまま） [ ] 1 Image Recognition - all models (2.2 GB) [ ] 2 > AlexNet (244 MB) [*] 3 > GoogleNet (54 MB) [ ] 4 > GoogleNet-12 (42 MB) [*] 5 > ResNet-18 (47 MB) [ ] 6 > ResNet-50 (102 MB) [ ] 7 > ResNet-101 (179 MB) [ ] 8 > ResNet-152 (242 MB) [ ] 9 > VGG-16 (554 MB) [ ] 10 > VGG-19 (575 MB) [ ] 11 > Inception-V4 (172 MB) [ ] 12 Object Detection - all models (395 MB) [ ] 13 > SSD-Mobilenet-v1 (27 MB) [*] 14 > SSD-Mobilenet-v2 (68 MB) [ ] 15 > SSD-Inception-v2 (100 MB) [*] 16 > PedNet (30 MB) [ ] 17 > NultiPed (30 MB) [*] 18 > FaceNet (24 MB) [*] 19 > DetectNet-COCO-Dog (29 MB) [ ] 20 > DetectNet-COCO-Bottle (29 MB) [ ] 21 > DetectNet-COCO-Chair (29 MB) [ ] 22 > DetectNet-COCO-Airplane (29 MB) [ ] 23 Semantic Segmentation - all (518 MB) [*] 24 > FCN-ResNet18-Cityscapes-512x256 (47 MB) [*] 25 > FCN-ResNet18-Cityscapes-1024x512 (47 MB) [ ] 26 > FCN-ResNet18-Cityscapes-2048x1024 (47 MB) [*] 27 > FCN-ResNet18-DeepScene-576x320 (47 MB) [ ] 28 > FCN-ResNet18-DeepScene-864x480 (47 MB) [*] 29 > FCN-ResNet18-MHP-512x320 (47 MB) [ ] 30 > FCN-ResNet18-MHP-512x320 (47 MB) [*] 31 > FCN-ResNet18-Pascal-VOC-320x320 (47 MB) [ ] 32 > FCN-ResNet18-Pascal-VOC-512x320 (47 MB) [*] 33 > FCN-ResNet18-SUN-RGBD-512x400 (47 MB) [ ] 34 > FCN-ResNet18-SUN-RGBD-640x512 (47 MB) [ ] 35 Semantic SEgmentation -legacy (1.4 GB) [ ] 36 > FCB-Alexnet-Cityscapes-SD (235 MB) [ ] 37 > FCB-Alexnet-Cityscapes-HD (235 MB) [ ] 38 > FCB-Alexnet-Aerial-FPV (7 MB) [ ] 39 > FCB-Alexnet-Pascal-VOC (235 MB) [ ] 40 > FCB-Alexnet-Synthia-CVPR (235 MB) [ ] 41 > FCB-Alexnet-Synthia-Summer-SD (235 MB) [ ] 42 > FCB-Alexnet-Synthia-Summer-HD (235 MB) [ ] 43 Image Processing - all models (138 MB) [ ] 44 > Deep-Homography-COCO (137 MB) [ ] 45 > Super-Resolution-BSD500 (1MB MB) → 了解
↓
PyTorchはスキップしても良いかも？
[ ] 1 PyTorch v1.1.0 for Python 2.7 [ ] 2 PyTorch v1.1.0 for Pythin 3.6 → 了解
↓ ビルド＆インストール＆ライブラリ登録
make sudo make install sudo ldconfig
動作確認
1回目（初回はいろいろ準備するものがあるようで時間がかかる）
cd $HOME/jetson-inference/build/aarch64/bin/ ./detectnet-console.py images/peds_0.jpg output_0.jpg
結果
↓
2回目
./detectnet-console.py images/peds_1.jpg output_1.jpg
結果
↓
参考
www.youtube.com jetson-inference/building-repo-2.md at master · dusty-nv/jetson-inference · GitHub
補足
# ダウンロードツール起動コマンド cd jetson-inference/tools ./download-models.sh # PyTorchのインストーラー起動コマンド cd jetson-inference/build ./install-pytorch.sh # 自前マシンへコピー scp jetson@jetson-desktop.local:/home/jetson/jetson-inference/build/aarch64/bin/images/peds_0.jpg $HOME/work/peds_0.jpg scp jetson@jetson-desktop.local:/home/jetson/jetson-inference/build/aarch64/bin/output_0.jpg $HOME/work/output_0.jpg scp jetson@jetson-desktop.local:/home/jetson/jetson-inference/build/aarch64/bin/images/peds_1.jpg $HOME/work/peds_1.jpg scp jetson@jetson-desktop.local:/home/jetson/jetson-inference/build/aarch64/bin/output_1.jpg $HOME/work/output_1.jpg
