maxbbraun / flir-adas-faces

Face bounding boxes for the FLIR ADAS Thermal Dataset

FLIR ADAS Faces

Face bounding boxes for the FLIR ADAS Thermal Dataset.

License

Only the annotation data (bounding-boxes.csv) is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License license.

The original terms and conditions of the FLIR ADAS Terms of Use apply to the images.

Method

1. Follow the instructions to download the FLIR_ADAS_1_3.tar.* files, then extract them:

FLIR_ADAS_DIR="flir-adas-database"
mkdir $FLIR_ADAS_DIR
tar -C $FLIR_ADAS_DIR -xvf FLIR_ADAS_1_3.tar.001 --strip-components=1

2. Filter the dataset down to images with people.

TMP_FLIR_ADAS_DIR="/tmp/$FLIR_ADAS_DIR"
TRAIN_CSV="train-people.csv"
VAL_CSV="val-people.csv"
VIDEO_CSV="video-people.csv"

python3 -m venv venv
. venv/bin/activate
pip install -r requirements.txt

python flir_convert.py \
  --input_dir=$FLIR_ADAS_DIR/train \
  --output_dir=$TMP_FLIR_ADAS_DIR/train \
  --output_csv=$TRAIN_CSV
python flir_convert.py \
  --input_dir=$FLIR_ADAS_DIR/val \
  --output_dir=$TMP_FLIR_ADAS_DIR/val \
  --output_csv=$VAL_CSV
python flir_convert.py \
  --input_dir=$FLIR_ADAS_DIR/video \
  --output_dir=$TMP_FLIR_ADAS_DIR/video \
  --output_csv=$VIDEO_CSV

FLIR_ADAS_PEOPLE_CSV="flir-adas-people.csv"
rm -f $FLIR_ADAS_PEOPLE_CSV
cat $TRAIN_CSV >> $FLIR_ADAS_PEOPLE_CSV
cat $VAL_CSV >> $FLIR_ADAS_PEOPLE_CSV
cat $VIDEO_CSV >> $FLIR_ADAS_PEOPLE_CSV

3. Upload the images to Cloud Storage.

FLIR_ADAS_BUCKET="gs://$FLIR_ADAS_DIR"
LOCATION="us-central1"
gsutil mb -l $LOCATION $FLIR_ADAS_BUCKET
gsutil iam ch allUsers:objectViewer $FLIR_ADAS_BUCKET

gsutil -m rsync -r $TMP_FLIR_ADAS_DIR $FLIR_ADAS_BUCKET
sed -E -i '' "s#$TMP_FLIR_ADAS_DIR#$FLIR_ADAS_BUCKET#" $FLIR_ADAS_PEOPLE_CSV
gsutil cp $FLIR_ADAS_PEOPLE_CSV $FLIR_ADAS_BUCKET

4. Set up a callback server using Node.js and MongoDB on Heroku.

5. Create the Scale annotation tasks:

SCALE_API_KEY=""  # Insert "Live API Key" from Scale dashboard.
SCALE_PROJECT=""  # Insert project name from Scale dashboard.
CALLBACK_URL=""  # Insert callback server URL from step 4.
INSTRUCTION="Draw a box around each face. The top should be above the forehead. The bottom should be below the chin. The left and right should span the width of the face, ignoring ears."

for f in $(cat $FLIR_ADAS_PEOPLE_CSV)
do
  path=$(printf '%s\n' "${f//$FLIR_ADAS_BUCKET\//}")
  url="https://storage.googleapis.com/$FLIR_ADAS_DIR/$path"
  curl "https://api.scale.com/v1/task/annotation" \
    -u "$SCALE_API_KEY:" \
    -d callback_url="$CALLBACK_URL" \
    -d instruction="$INSTRUCTION" \
    -d attachment_type=image \
    -d attachment="$url" \
    -d objects_to_annotate[0]="face" \
    -d with_labels=false \
    -d project="$SCALE_PROJECT"
done

6. Check the Scale Dashboard for all annotation tasks to be completed.

7. Export the annotations as a CSV file:

MONGO_HOSTNAME=""  # Insert database host from Heroku/mLab dashboard.
MONGO_DATABASE=""  # Insert database name from Heroku/mLab dashboard.
MONGO_USERNAME=""  # Insert username from Heroku/mLab dashboard.
MONGO_PASSWORD=""  # Insert password from Heroku/mLab dashboard.
MONGO_COLLECTION="tasks"
BBOXES_CSV="bounding-boxes.csv"

python mongo_export.py \
  --hostname=$MONGO_HOSTNAME \
  --database=$MONGO_DATABASE \
  --collection=$MONGO_COLLECTION \
  --username=$MONGO_USERNAME \
  --password=$MONGO_PASSWORD \
  --output_csv=$BBOXES_CSV

8. Clean up

rm FLIR_ADAS_1_3.tar.*
rm -rf $FLIR_ADAS_DIR
rm -rf $TMP_FLIR_ADAS_DIR
rm *-people.csv
gsutil rm -r $FLIR_ADAS_BUCKET