Skip to content

Redis

Requirements#

Requires the image-dataset-converter-redis library.

Object detection example#

In this example we are using a prebuilt yolov5 model (using MS-COCO) to make predictions on the frames that come from a dashcam video, overlay the predictions on the images and display them. For the model we will be using an existing docker container.

Requirements#

NB: No GPU required.

Additional image-dataset-converter libraries:

Data#

Input#

dashcam raw

Output#

dashcam annotated

Preparation#

NB: Place all the downloads in the current directory

  • Download the dashcam01.mp4 video from the BoofCV project
  • Download the yolo5n.pt model
  • Download the coco.yaml data description for the yolo5n model

  • The host machine must have a Redis server instance running. Two options:

    1. Install it natively via sudo apt-get install redis (and then restart it with sudo systemctl restart redis)
    2. Spin up a docker container with: docker run --net=host --name redis-server -d redis

Yolov5 model#

The following command launches a Yolov5 model via the container's yolov5_predict_redis command, running on the CPU: 

docker run \
    --net=host -u $(id -u):$(id -g) \
    -v `pwd`:/workspace \
    -it waikatodatamining/pytorch-yolov5:2022-01-21_cpu \
  yolov5_predict_redis \
    --redis_in images \
    --redis_out predictions \
    --model /workspace/yolov5n.pt \
    --data /workspace/coco.yaml

image-dataset-converter (direct prediction)#

The following pipeline loads every 2nd frame from the dashcam01.mp4 video, obtains predictions from the Yolov5 model (using the Redis backend), overlays the predictions and then displays them:

idc-convert \
  -l INFO \
  from-video-file \
    -i ./dashcam01.mp4 \
    -n 2 \
    -t od \
  redis-predict-od \
    --channel_out images \
    --channel_in predictions \
    --timeout 1.0 \
  add-annotation-overlay-od \
    --outline_alpha 255 \
    --outline_thickness 1 \
    --fill \
    --fill_alpha 128 \
    --vary_colors \
    --font_size 10 \
    --text_format "{label}: {score}" \
    --text_placement T,L \
    --force_bbox \
  image-viewer \
    --size 800,224 \
    --delay 1

image-dataset-converter (via meta-sub-images)#

Like the above example, but uses the meta-sub-images filter to split the incoming images into two and sends them to the model one-by-one, then assembles the predictions and forwards them:

idc-convert \
  -l INFO \
  from-video-file \
    -i ./dashcam01.mp4 \
    -n 2 \
    -t od \
  meta-sub-images \
    -l INFO \
    -r 0,0,400,224 400,0,400,224 \
    --merge_adjacent_polygons \
    --base_filter "redis-predict-od --channel_out images --channel_in predictions --timeout 1.0" \
  add-annotation-overlay-od \
    --outline_alpha 255 \
    --outline_thickness 1 \
    --fill \
    --fill_alpha 128 \
    --vary_colors \
    --font_size 10 \
    --text_format "{label}: {score}" \
    --text_placement T,L \
    --force_bbox \
  image-viewer \
    --size 800,224 \
    --delay 1

NB:

  • Requires the image-dataset-converter-imgaug library.
  • Meta-data of merged polygons gets lost apart from the label (type), which has to be the same, and any score values, which get averaged.