A docker image for easier deployment of the hawkweed yolo detection model.
- 202301motiur
- Introduction
- Getting a copy of this repo to your computer
- Build the container.
- Quickstart
- Using the python script directly for image inference
- Acknowledgements
The docker container allows deployment of the model without having to deal with tedious software installation steps and operating system differences. The image is based on Ubuntu linux 20.04 with the GPU CUDA drivers pre-installed. The high resolution data processed by this model require GPU with a reasonably large amount of RAM (>8GB), however, the detections can be carried out in CPU mode (~12 seconds per image).
The detect.py script that runs the model inference can be run using the docker run command or interactively by logging into the docker container itself after it is started in -d (detected mode). Instructions for both modes of interaction are shown below.
The following steps are written for windows users, see the Linux branch if you use that system. I've written a somewhat user-friendly guide to installing docker on windows. This guide includes an example running a camera trap detection model (megadetector) which you might want to try first if you are new to all of this.
click on the code button on the main page of the repo (see below) and download
the zip file to your desktop (or a location of your choice) and unzip the folder.
Note these steps will take some time depending on the speed of your internet connection.
Copy the yolov5 directory (request from Motiur) to this directory, it should look like this
Then copy the two test scripts test-cpu.sh and test-gpu.sh into the yolov5 directory.
Open a command prompt in the repo directory by typing cmd in the browser window (circled below).
This should open a command line prompt in the current directory.
Type (or copy/paste) the following command to build the container.
docker build -f Dockerfile . -t dwheelerau/hawkweed:ubuntu2004
Docker will start to download the entire internet (so this may take some time)!
If everything goes well you should see "FINISHED" printed to screen.
The following should mount your current directory and save the test results to it.
docker run -it -v %cd%:/project dwheelerau/hawkweed:ubuntu2004 /bin/bash -c "cd /project && bash /build/202301motiur/yolov5/test-cpu.sh"
If everything works the following should print to screen.
You will need a reasonably large RAM GPU.
docker run --gpus all -it -v %cd%:/project dwheelerau/hawkweed:ubuntu2004 /bin/bash -c "cd /project && bash /build/202301motiur/yolov5/test-gpu.sh"
The following provides several ways of using the model for hawk weed inference.
This assumes you have your images in a folder called images located in your current working directory.
Start the container in detached mode -d after mounting your cwd.
# -it is interactive tty
# -d detached so keep running after command is executed
docker run -it -d -v %cd%:/project dwheelerau/hawkweed:ubuntu2004
See what it is called.
docker ps -a
CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES
04f40d0ae645 dwheelerau/hawkweed:ubuntu2004 "/bin/bash" 7 seconds ago Up 6 seconds dazzling_moser
We can see the container is called dazzling_moser (this will be different for you). Take note of this name as we use it in the command below.
Execute a command on the running container (the projects directory will already be mounted in your cwd). Change the --source location if your images
are stored somewhere other than a folder called images. It will save
the detections to a new folder called images_out.
# change the name based on the above output from `ps -a`
# if GPU available: --gpus all
# if GPU available: --device gpu
docker exec -it dazzling_moser /bin/bash -c "cd /project && python /build/202301motiur/yolov5/detect.py --device cpu --weights /build/202301motiur/yolov5/runs/train/exp72/weights/best.pt --img 5320 7968 --conf 0.45 --iou 0.35 --source images --name images_out --project /project/"
A new folder is created called images_out with the results.
Start the container in detached mode after mounting your cwd.
# -it is interactive tty
# -d detached so keep running after command is executed
docker run -it -d -v `pwd`:/project dwheelerau/hawkweed:ubuntu2004
See what it is called (see last column of the NAMES column at the far left of the printout below.
docker ps -a
##CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES
##2414a542a929 dwheelerau/hawkweed:ubuntu2004 "/bin/bash" 4 seconds ago Up 3 seconds determined_haibt
Now you can log in. To see the full list of command line options use -h.
docker exec -it determined_haibt /bin/bash
python ./202301motiur/yolov5/detect.py -h
usage: detect.py [-h] [--weights WEIGHTS [WEIGHTS ...]] [--source SOURCE]
[--data DATA] [--imgsz IMGSZ [IMGSZ ...]]
[--conf-thres CONF_THRES] [--iou-thres IOU_THRES]
[--max-det MAX_DET] [--device DEVICE] [--view-img]
[--save-txt] [--save-conf] [--save-crop] [--nosave]
[--classes CLASSES [CLASSES ...]] [--agnostic-nms]
[--augment] [--visualize] [--update] [--project PROJECT]
[--name NAME] [--exist-ok] [--line-thickness LINE_THICKNESS]
[--hide-labels] [--hide-conf] [--half] [--dnn]
optional arguments:
-h, --help show this help message and exit
--weights WEIGHTS [WEIGHTS ...]
model path(s)
--source SOURCE file/dir/URL/glob, 0 for webcam
--data DATA (optional) dataset.yaml path
--imgsz IMGSZ [IMGSZ ...], --img IMGSZ [IMGSZ ...], --img-size IMGSZ [IMGSZ ...]
inference size h,w
--conf-thres CONF_THRES
confidence threshold
--iou-thres IOU_THRES
NMS IoU threshold
--max-det MAX_DET maximum detections per image
--device DEVICE cuda device, i.e. 0 or 0,1,2,3 or cpu
--view-img show results
--save-txt save results to *.txt
--save-conf save confidences in --save-txt labels
--save-crop save cropped prediction boxes
--nosave do not save images/videos
--classes CLASSES [CLASSES ...]
filter by class: --classes 0, or --classes 0 2 3
--agnostic-nms class-agnostic NMS
--augment augmented inference
--visualize visualize features
--update update all models
--project PROJECT save results to project/detect,name
--name NAME save results to project/name
--exist-ok existing project/name ok, do not increment
--line-thickness LINE_THICKNESS
bounding box thickness (pixels)
--hide-labels hide labels
--hide-conf hide confidences
--half use FP16 half-precision inference
--dnn use OpenCV DNN for ONNX inference
You should be able to interact with this container via the command line and /build/ is mounted in your cwd on the host.
Motiur - model development
Nico - docker insights
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent