Yolov5_DeepSort_Pytorch
製品情報
ユースケース
公開チャット
サポートプラン
現在利用可能なOSSプランはありません
リポジトリのプロバイダーまたは寄稿者である場合は、OSSプランの追加を開始できます。
OSSプランを追加するこのOSSプランをお探しの場合は、お問い合わせください。
製品詳細
Easy training program of Yolov5 Deep Sort with PyTorch
- This repository is folked from https://github.com/mikel-brostrom/Yolov5_DeepSort_Pytorch
- Some explanation of this repository is here(In Japanese)
Prerequisites
- Already installed Docker
How to use
-
create environment (common step)
- execute following script in command line
# clone this repository git clone --recurse-submodules https://github.com/tmfi-analytics/Yolov5_DeepSort_Pytorch.git
build Docker container using Dockerfile
cd Yolov5_DeepSort_Pytorch docker build -t [YOUR CONTAINER NAME] .
- execute following script in command line
3 ways to use
-
Just execute human tracking demo
-
edit config.sh as following
# 訓練済みのモデルで物体追跡をする場合はここを1に設定 JUST_PREDICTION=1 # YOLOV5の訓練に自前のデータセットを使用する場合はここを1に設定 USE_CUSTOM_DATASET=0
- execute docker run command
cd Yolov5_DeepSort_Pytorch docker run -it --env-file=config.sh -v `pwd`/data:/app/data --shm-size=2048m [YOUR CONTAINER NAME]
- watch the output video
ls data/result
-
-
train the human detection model and track using prepared dataset
-
edit config.sh as following
# 訓練済みのモデルで物体追跡をする場合はここを1に設定 JUST_PREDICTION=0 # YOLOV5の訓練に自前のデータセットを使用する場合はここを1に設定 USE_CUSTOM_DATASET=0
- execute docker run command
cd Yolov5_DeepSort_Pytorch docker run -it --env-file=config.sh -v `pwd`/data:/app/data --shm-size=2048m [YOUR CONTAINER NAME]
-
watch the output video
ls data/result
The movies for train/val are downloaded by open image dataset v6
https://storage.googleapis.com/openimages/web/index.html
The movie for test is downloaded by following link
https://www.motionelements.com/ja/stock-video-12530972-shopping-mall-in-japan?query_id=solr_40d80e54990840ae85da7fcfb8008d67&position=5
-
-
train the human detection model and track using custom dataset
- put train and val dataset as following
put image files in images directory and label files in labels directory. the pair of image file and label file must be same file name
please refer the following link about label format
https://github.com/ultralytics/yolov5/wiki/Train-Custom-Data - put mp4 file for test as data/test/test.mp4
-
edit config.sh as following
# 訓練済みのモデルで物体追跡をする場合はここを1に設定 JUST_PREDICTION=0 # YOLOV5の訓練に自前のデータセットを使用する場合はここを1に設定 USE_CUSTOM_DATASET=1
- execute docker run command
cd Yolov5_DeepSort_Pytorch docker run -it --env-file=config.sh -v `pwd`/data:/app/data --shm-size=2048m [YOUR CONTAINER NAME]
- watch the output video
ls data/result
- put train and val dataset as following