demo.bash

#!/bin/bash

DATE=$(date "+%Y%m%d_%H%M%S")
GPU_ID=$1
VIDEO_FULLNAME=$(realpath "$2")
VIDEO_OUTPUTDIR=$3
: ${VIDEO_OUTPUTDIR:="outputs"}

VIDEO_DIRNAME=$(dirname "$VIDEO_FULLNAME")
VIDEO_BASENAME=$(basename "$VIDEO_FULLNAME")
VIDEO_EXT=${VIDEO_BASENAME##*.}
VIDEO_NAME=${VIDEO_BASENAME%.*}

EXPECT_DETECT_OUTPUT="$VIDEO_OUTPUTDIR/$VIDEO_NAME/$VIDEO_NAME"_detection.txt
EXPECT_CLASS_OUTPUT="$VIDEO_OUTPUTDIR/$VIDEO_NAME/$VIDEO_NAME"_ensemble.txt
EXPECT_MOUSE_OUTPUT="$VIDEO_OUTPUTDIR/$VIDEO_NAME/$VIDEO_NAME"_mouse.json
EXPECT_PATHS_OUTPUT="$VIDEO_OUTPUTDIR/$VIDEO_NAME/paths.csv"

# EXPECT_OBSERVER_FILE="$VIDEO_DIRNAME/$VIDEO_NAME"_observer.csv
# EXPECT_ACTION_OUTPUT="output/path/$VIDEO_NAME/action_paths.csv"
# EXPECT_ACTION_CLIPS_OUTPUT="$VIDEO_DIRNAME/clips"

# step1: inference detection model
if [[ ! -e $EXPECT_DETECT_OUTPUT ]];then
    echo "$EXPECT_DETECT_OUTPUT not exist... processing"
    cd detection
    source $(pipenv --venv)/bin/activate
    python generate_bbox.py \
    --gpu $GPU_ID \
    --weights ../models/detection/lstm_resnet_beetle_rezoom/save.ckpt-1300000 \
    --video-root $VIDEO_DIRNAME \
    --video-type $VIDEO_EXT \
    --output-dir $VIDEO_OUTPUTDIR
    deactivate && cd ..
fi

# step2: inference classification model
if [[ -e $EXPECT_DETECT_OUTPUT && ! -e $EXPECT_CLASS_OUTPUT ]]; then
    echo "$EXPECT_CLASS_OUTPUT not exist... processing"
    source $(pipenv --venv)/bin/activate
    python3 ensemble_predicts.py \
    --gpu $GPU_ID \
    --video "$VIDEO_FULLNAME" \
    --input "$EXPECT_DETECT_OUTPUT" \
    --models models/classification/resnet.h5 models/classification/xception.h5
    deactivate
fi 

# step3: apply tracking algorithm
if [[ -e $EXPECT_DETECT_OUTPUT && \
      -e $EXPECT_CLASS_OUTPUT && \
      ! -e $EXPECT_PATHS_OUTPUT ]]; then
    echo "$EXPECT_PATHS_OUTPUT not exist... processing"
    source $(pipenv --venv)/bin/activate
    python3 main.py \
    --video "$VIDEO_FULLNAME" \
    --input "$EXPECT_CLASS_OUTPUT" \
    --config config/default.json \
    --no-show-video --no-save-video
    deactivate
fi

# # optional: generate action data or predict action
# if [ "$4" = "action_data" ];then
#     source $(pipenv --venv)/bin/activate
#     # convert observer
#     if [[ -e $EXPECT_OBSERVER_FILE && \
#           -e $EXPECT_PATHS_OUTPUT && \
#           ! -e $EXPECT_ACTION_OUTPUT ]]; then
#         echo "Convert observer file..."

#         python3 action/convert_observer.py \
#         -v $VIDEO_FULLNAME \
#         -i $EXPECT_OBSERVER_FILE \
#         -p $EXPECT_PATHS_OUTPUT
#     fi

#     # generate video clips fro action training data
#     if [[ -e $EXPECT_ACTION_OUTPUT && ! -e $EXPECT_ACTION_CLIPS_OUTPUT ]]; then
#         echo "Clips video for action training data..."
#         python3 action/generate_video_clips.py \
#         -v $VIDEO_FULLNAME \
#         -i $EXPECT_ACTION_OUTPUT
#     fi
#     deactivate
# fi