abcli/image_classifier.sh

#! /usr/bin/env bash

function image_classifier() {
    abcli_image_classifier $@
}

function abcli_image_classifier() {
    local task=$(abcli_unpack_keyword "$1" help)

    if [ "$task" == "help" ] ; then
        abcli_help_line "$abcli_cli_name image_classifier install" \
            "install image_classifier."
        abcli_help_line "$abcli_cli_name image_classifier list [object_1] [model=object/*saved]" \
            "list [saved/object model object_1]."
        abcli_help_line "$abcli_cli_name image_classifier predict data_1 [name_1] [data=filename/*object/url,model=object/*saved]" \
            "run fashion_mnist saved/object model name_1 predict on filename/object/url data_1."
        abcli_help_line "$abcli_cli_name image_classifier save [name_1] [object_1] [force]" \
            "[force] save image_classifier [in object_1] [as name_1]."
        abcli_help_line "$abcli_cli_name image_classifier train object_1" \
            "train image_classifier on data object_1."

        if [ "$(abcli_keyword_is $2 verbose)" == true ] ; then
            python3 -m image_classifier --help
        fi
        return
    fi

    if [[ $(type -t abcli_image_classifier_$task) == "function" ]] ; then
        abcli_image_classifier_$task ${@:2}
        return
    fi

    if [ "$task" == "install" ] ; then
        conda install -y -c anaconda seaborn
        return
    fi

    if [ "$task" == "list" ] ; then
        local model_name=$2

        if [ -z "$model_name" ] ; then
            ls $abcli_path_git/image-classifier/saved_model
            return
        fi

        local options=$3
        local model_source=$(abcli_option "$options" "model" saved)
        local do_browse=$(abcli_option_get_unpacked "$options" "browser" 0)

        local model_path=$(abcli_huggingface get_model_path image-classifier "$model_name" "$options")

        if [ "$model_source" == "object" ] ; then
            local model_object=$(python3 -c "print('$model_path'.split('/')[-1])")
            abcli_download object $model_object
        fi

        python3 -m image_classifier \
            list \
            --model_path $model_path \
            ${@:4}

        if [ "$do_browse" == 1 ] && [ "$model_source" == "object" ] ; then
            abcli_browser $model_object
        fi

        return
    fi

    if [ "$task" == "save" ] ; then
        abcli_huggingface save \
            image-classifier \
            $(abcli_clarify_arg "$2" image-classifier) \
            ${@:3}
        return
    fi

    abcli_log_error "-fashion_mnist: image-classifier: $task: command not found."
}

function abcli_image_classifier_predict() {
    local data_object=$(abcli_clarify_object "$1")

    local model_name=$2

    local options=$3
    local data_source=$(abcli_option "$options" "data" object)
    local model_source=$(abcli_option "$options" "model" saved)

    if [ "$(abcli_keyword_is $data_object validate)" == true ] ; then
        if [ "$data_source" == "object" ] ; then
            abcli_log_error "-imge-classifier: predict: validation object not found."
            return
        fi

        local data_object="https://upload.wikimedia.org/wikipedia/commons/thumb/8/8b/Claquettes-peto.jpg/1024px-Claquettes-peto.jpg"
        local data_source="url"
    fi

    if [ "$data_source" == "object" ] ; then
        abcli_download object $data_object
    fi


    local model_path=$(abcli_huggingface get_model_path image-classifier "$model_name" "$options")

    if [ "$model_source" == "object" ] ; then
        local model_object=$(python3 -c "print('$model_path'.split('/')[-1])")
        abcli_download object $model_object
    fi

    abcli_log "image_classifier($model_path).predict($data_object): $options"

    if [ ! -f "$abcli_object_root/$data_object/test_images.pyndarray" ] && [ "$data_source" == "object" ] ; then
        python3 -m image_classifier \
            preprocess \
            --infer_annotation 0 \
            --model_path $model_path \
            --objects $abcli_object_root/$data_object \
            --output_path $abcli_object_root/$data_object \
            --purpose predict \
            ${@:4}
    fi

    if [ "$data_source" == "object" ] ; then
        cp -v $abcli_object_root/$data_object/*.pyndarray .
        cp -v $model_path/image_classifier/model/class_names.json .

        python3 -m image_classifier \
            predict \
            --data_path $abcli_object_root/$data_object \
            --model_path $model_path \
            --output_path $abcli_object_path \
            ${@:4}

        abcli_tag set . image_classifier,predict
    else
        local is_url=0
        if [ "$data_source" == "url" ] ; then
            local is_url=1
        fi

        python3 -m image_classifier \
            predict_image \
            --data_path $data_object \
            --is_url $is_url \
            --model_path $model_path \
            --output_path $abcli_object_path \
            ${@:4}  
    fi
}

function abcli_image_classifier_train() {
    local data_object=$(abcli_clarify_object "$1" $abcli_object_name)

    abcli_download object $data_object

    local options=$2
    local do_color=$(abcli_option_int "$options" "color" 0)
    local do_convnet=$(abcli_option_int "$options" "convnet" 0)
    local do_validate=$(abcli_option_int "$options" "validate" 0)

    local extra_args=""
    if [ "$do_validate" == 1 ] ; then
        local extra_args="--epochs 2"
    fi

    abcli_log "image_classifier.train($data_object): $options"

    python3 -m image_classifier \
        train \
        --color $do_color \
        --convnet $do_convnet \
        --data_path $abcli_object_root/$data_object \
        --model_path $abcli_object_path \
        $extra_args \
        ${@:3}

    abcli_tag set . image_classifier,train
}