#確認裝置空間 至少要20G 最好裝D槽

1.確認顯卡唯一碼

在終端輸入
nvidia-smi -L

Image Not Showing Possible Reasons
  • The image was uploaded to a note which you don't have access to
  • The note which the image was originally uploaded to has been deleted
Learn More →

複製括號內的UUID,貼到下面git中lock/uuid.lock的位置

2.設定環境變數後編譯

需確認系統才能編譯(Linux/Windows/Jetson/Ubuntu)
需編譯的 git https://leda-repo.seadeep.ai/LEDA/seadeep-next/src/branch/master/

Windows Powershell

$env:CGO_ENABLED="0"

#GOOS依照客戶端的系統去選擇,如果雙方的系統一樣就不用輸入
$env:GOOS="linux"
$env:GOOS="windows"

bash

set %CGO_ENABLED%="0"

#GOOS依照客戶端的系統去選擇,如果雙方的系統一樣就不用輸入
set %GOOS%="linux"
set %GOOS%="windows"

Ubuntu Terminal

CGO_ENABLED="0"
set %CGO_ENABLED%="0"

#GOOS依照客戶端的系統去選擇,如果雙方的系統一樣就不用輸入
GOOS="linux"
GOOS="windows"

輸入完後可以用go env確認CGO_ENABLED、GOOS

設定好後就可以編譯了
go build -v ./cmd/lite
編譯好會產生 lite.exe (Seadeep Server)

3. 準備核心 (在Seadeep-Lite下面)

需準備的 git https://leda-repo.seadeep.ai/LEDA/seadeep-lite
先確認環境在python 3.8
python -V
隱藏原始碼
python .\compile.py
可刪除 kernel/{object_detection/classification/segmentation} 資料夾

產物: 整包 seadeep-lite 資料夾 (裡面都是pyc) (模型參數) 丟給客戶

4. 打包UI

cd earth
yarn build-only-lite

*備註:vite 版本要為 4.0.0 - 5.0.0 (如果沒有vite可以輸入以下指令)

yarn add vite@^5.0.0

打包完會出現dist 資料夾(UI)

以下流程可以由實習生操作

客戶端

1.解壓縮Seadeep-Lite壓縮檔

2.把lite.exe放入解壓縮完的資料夾裡

3.安裝python環境

先確認Cuda版本
nvidia-smi

Image Not Showing Possible Reasons
  • The image was uploaded to a note which you don't have access to
  • The note which the image was originally uploaded to has been deleted
Learn More →

建立conda環境

conda create -n py38 python=3.8
conda activate py38

到pytorch網站找適合的版本下載
https://pytorch.org/get-started/previous-versions/#v191
CUDA版本在11以上的話就選擇11.1那個版本

Image Not Showing Possible Reasons
  • The image was uploaded to a note which you don't have access to
  • The note which the image was originally uploaded to has been deleted
Learn More →

pip install torch==1.9.1+cu111 torchvision==0.10.1+cu111 torchaudio==0.9.1 -f https://download.pytorch.org/whl/torch_stable.html

確認torch是否安裝成功

python
import torch
torch.cuda.is_available()

顯示True代表成功

Image Not Showing Possible Reasons
  • The image was uploaded to a note which you don't have access to
  • The note which the image was originally uploaded to has been deleted
Learn More →

在建立好的conda裡進入Seadeep-Lite的資料夾

pip install -r requirements_kernal.txt

4.建立捷徑到桌面,設定圖示

先用記事本打開run.bat確認裡面的conda環境名稱正確
右鍵run.bat建立捷徑 然後丟到桌面或想放置的地方
接著右鍵run.bat-捷徑->點選內容->選擇變更圖示->瀏覽選擇圖示seadeep.ico
最後將名稱改成SeaDeep 雙擊就可以執行了
執行後在瀏覽器輸入localhost:9816就可以進入SeaDeep的網站

5.測試SeaDeep四大核心(od ae cls seg)正常運行

跑訓練的時候可以減少學習次數 才不會花太多時間測試
訓練、推論都要測試過一次

補充 客戶設備若不能聯網 需要自行下載權重到客戶端

ERROR:

遇到問題可以先檢查train.bat infer.bat(windows)

train.sh infer.sh(linux)的內容有沒有正確

numpy版本過低:

Image Not Showing Possible Reasons
  • The image was uploaded to a note which you don't have access to
  • The note which the image was originally uploaded to has been deleted
Learn More →

升級numpy
pip install numpy==1.24.4

image
在Seadeep-Lite資料夾裡找到infer.bat 用記事本打開
原本是infer.py 改成infer.pyc 因為已經編譯過了