Skip to content

Inference

Goal

  • Build the container image
  • Run the tile based classification inference in the container image tagged localhost/inference:latest.

Lab

This step has a dedicated lab available at /workspace/inference-eoap/practice-labs/Container.ipynb

The container recipe

The tile based classification inference has a recipe to build the container image.

The inference step container image recipe is:

inference/Dockerfile
1

Building the container

Build the container images with:

terminal
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
export WORKSPACE=/workspace/inference-eoap

command -v podman >/dev/null 2>&1 && { 

    podman build --format docker -t localhost/inference:latest ${WORKSPACE}/tile-based-classification/command-line-tools/inference

} || command -v docker >/dev/null 2>&1 && { 

    docker build -t localhost/inference:latest ${WORKSPACE}/tile-based-classification/command-line-tools/inference
}

Expected outcome

The local container registry lists the built images:

(base) jovyan@coder-mrossi:~/runs$ podman images | grep localhost
localhost/inference  latest      534f3f10c06e  11 minutes ago  530 MB

How to run a step in a container

We'll use podman container engine (docker is also fine).

The command to run the inference step in the container is:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
WORKSPACE=/workspace/inference-eoap
RUNTIME=${WORKSPACE}/runs
mkdir -p ${RUNTIME}
cd ${RUNTIME}

# check if the results.json file exists
if [ ! -f ${WORKSPACE}/runs/results.json ]; then
    echo "results.json file not found, run the stage-and-cog.cwl workflow first."
    exit 1
fi

podman run \
    -i \
    --userns=keep-id \
    --mount=type=bind,source=/workspace/inference-eoap/runs,target=/runs \
    --workdir=/runs \
    --read-only=true \
    --user=1001:100 \
    --rm \
    --env=HOME=/runs \
    --env=PYTHONPATH=/app \
    localhost/inference:latest \
    python \
    -m \
    app \
    --input-item \
    $( cat ${WORKSPACE}/runs/results.json | jq -r .stac_catalog.path )

Let's break down what this command does:

  • podman run: This is the command to run a container.
  • -i: This flag makes the container interactive, allowing you to interact with it via the terminal.
  • --userns=keep-id: It instructs podman to keep the user namespace ID.
  • --mount=type=bind,source=/workspace/runs,target=/runs: This option mounts a directory from the host system to the container. In this case, it mounts the /workspace/runs directory on the host to the /runs directory inside the container.
  • --workdir=/runs: Sets the working directory inside the container to /runs.
  • --read-only=true: Makes the file system inside the container read-only, meaning you can't write or modify files inside the container.
  • --user=1001:100: Specifies the user and group IDs to be used within the container.
  • --rm: This flag tells podman to remove the container after it has finished running.
  • --env=HOME=/runs: Sets the HOME environment variable inside the container to /runs.
  • --env=PYTHONPATH=/app: Sets the PYTHONPATH environment variable inside the container to /app.
  • localhost/inference:latest: This is the name of the container image that you want to run. It's pulling the image from the local container registry with the name "inference" and the "latest" tag.
  • python -m app: This is the command to run inside the container. It runs a Python module named "app".
  • --input-item ...: Specifies the path to the folder where there is a catalog.json file.

Expected outcome

The folder /workspace/inference-eoap/runs contains: 

(base) jovyan@jupyter-mrossi--training:~/inference-eoap$ tree runs/
runs
├── S2A_T33TVM_20240125T100359_L1C-classification
│   ├── S2A_T33TVM_20240125T100359_L1C-classification.json
│   └── classification.tif
├── catalog.json
├── results.json
└── ydmns4od
    ├── S2A_T33TVM_20240125T100359_L1C
    │   ├── S2A_T33TVM_20240125T100359_L1C.json
    │   ├── blue.tif
    │   ├── cirrus.tif
    │   ├── coastal.tif
    │   ├── green.tif
    │   ├── nir.tif
    │   ├── nir08.tif
    │   ├── nir09.tif
    │   ├── red.tif
    │   ├── rededge1.tif
    │   ├── rededge2.tif
    │   ├── rededge3.tif
    │   ├── swir16.tif
    │   └── swir22.tif
    └── catalog.json

3 directories, 19 files