tm_landmark_timvx.cpp

/*
 * Licensed to the Apache Software Foundation (ASF) under one
 * or more contributor license agreements.  See the NOTICE file
 * distributed with this work for additional information
 * regarding copyright ownership.  The ASF licenses this file
 * to you under the Apache License, Version 2.0 (the
 * License); you may not use this file except in compliance
 * with the License.  You may obtain a copy of the License at
 *
 *   http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing,
 * software distributed under the License is distributed on an
 * AS IS BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
 * KIND, either express or implied.  See the License for the
 * specific language governing permissions and limitations
 * under the License.
 */

/*
 * Copyright (c) 2020, OPEN AI LAB
 * Author: qtang@openailab.com
 */

#include <iostream>
#include <functional>

#include "common.h"
#include "tengine/c_api.h"
#include "tengine_operations.h"

#define DEFAULT_REPEAT_COUNT 1
#define DEFAULT_THREAD_COUNT 1

void get_input_uint8_data(const char* image_file, uint8_t* input_data, int img_h, int img_w, float* mean, float* scale,
                          float input_scale, int zero_point)
{
    image img = imread_process(image_file, img_w, img_h, mean, scale);

    float* image_data = (float*)img.data;

    for (int i = 0; i < img_w * img_h * 3; i++)
    {
        int udata = (round)(image_data[i] / input_scale + (float)zero_point);
        if (udata > 255)
            udata = 255;
        else if (udata < 0)
            udata = 0;

        input_data[i] = udata;
    }

    free_image(img);
}

void show_usage()
{
    fprintf(stderr, "[Usage]:  [-h]\n    [-m model_file] [-i image_file] [-r repeat_count] [-t thread_count]\n");
}

int main(int argc, char* argv[])
{
    int repeat_count = DEFAULT_REPEAT_COUNT;
    int num_thread = DEFAULT_THREAD_COUNT;
    char* model_file = nullptr;
    char* image_file = nullptr;
    int img_h = 144;
    int img_w = 144;
    float mean[3] = {128.f, 128.f, 128.f};
    float scale[3] = {0.0039, 0.0039, 0.0039};

    int res;
    while ((res = getopt(argc, argv, "m:i:r:t:h:")) != -1)
    {
        switch (res)
        {
        case 'm':
            model_file = optarg;
            break;
        case 'i':
            image_file = optarg;
            break;
        case 'r':
            repeat_count = atoi(optarg);
            break;
        case 't':
            num_thread = atoi(optarg);
            break;
        case 'h':
            show_usage();
            return 0;
        default:
            break;
        }
    }

    /* check files */
    if (model_file == nullptr)
    {
        fprintf(stderr, "Error: Tengine model file not specified!\n");
        show_usage();
        return -1;
    }

    if (image_file == nullptr)
    {
        fprintf(stderr, "Error: Image file not specified!\n");
        show_usage();
        return -1;
    }

    if (!check_file_exist(model_file) || !check_file_exist(image_file))
        return -1;

    /* set runtime options */
    struct options opt;
    opt.num_thread = num_thread;
    opt.cluster = TENGINE_CLUSTER_ALL;
    opt.precision = TENGINE_MODE_UINT8;
    opt.affinity = 0;

    /* inital tengine */
    if (init_tengine() != 0)
    {
        fprintf(stderr, "Initial tengine failed.\n");
        return -1;
    }
    fprintf(stderr, "tengine-lite library version: %s\n", get_tengine_version());

    /* create VeriSilicon TIM-VX backend */
    context_t timvx_context = create_context("timvx", 1);
    int rtt = set_context_device(timvx_context, "TIMVX", nullptr, 0);
    if (0 > rtt)
    {
        fprintf(stderr, " add_context_device VSI DEVICE failed.\n");
        return -1;
    }

    /* create graph, load tengine model xxx.tmfile */
    graph_t graph = create_graph(timvx_context, "tengine", model_file);
    if (NULL == graph)
    {
        fprintf(stderr, "Create graph failed.\n");
        return -1;
    }

    /* set the input shape to initial the graph, and prerun graph to infer shape */
    int img_size = img_h * img_w * 3;
    int dims[] = {1, 3, img_h, img_w}; // nchw
    uint8_t* input_data = (uint8_t*)malloc(img_size);

    tensor_t input_tensor = get_graph_input_tensor(graph, 0, 0);
    if (input_tensor == nullptr)
    {
        fprintf(stderr, "Get input tensor failed\n");
        return -1;
    }

    if (set_tensor_shape(input_tensor, dims, 4) < 0)
    {
        fprintf(stderr, "Set input tensor shape failed\n");
        return -1;
    }

    if (set_tensor_buffer(input_tensor, input_data, img_size) < 0)
    {
        fprintf(stderr, "Set input tensor buffer failed\n");
        return -1;
    }

    /* prerun graph, set work options(num_thread, cluster, precision) */
    if (prerun_graph_multithread(graph, opt) < 0)
    {
        fprintf(stderr, "Prerun multithread graph failed.\n");
        return -1;
    }

    /* prepare process input data, set the data mem to input tensor, quantize fp32 to uint8 */
    float input_scale = 0.f;
    int input_zero_point = 0;
    get_tensor_quant_param(input_tensor, &input_scale, &input_zero_point, 1);
    get_input_uint8_data(image_file, input_data, img_h, img_w, mean, scale, input_scale, input_zero_point);

    /* run graph */
    double min_time = DBL_MAX;
    double max_time = DBL_MIN;
    double total_time = 0.;
    for (int i = 0; i < repeat_count; i++)
    {
        double start = get_current_time();
        if (run_graph(graph, 1) < 0)
        {
            fprintf(stderr, "Run graph failed\n");
            return -1;
        }
        double end = get_current_time();
        double cur = end - start;
        total_time += cur;
        if (min_time > cur)
            min_time = cur;
        if (max_time < cur)
            max_time = cur;
    }
    printf("Repeat [%d] min %.3f ms, max %.3f ms, avg %.3f ms\n", repeat_count, min_time, max_time,
           total_time / repeat_count);

    /* get output tensor */
    tensor_t output_tensor = get_graph_output_tensor(graph, 0, 0);

    float output_scale = 0.f;
    int output_zp = 0;
    get_tensor_quant_param(output_tensor, &output_scale, &output_zp, 1);
    uint8_t* data = (uint8_t*)(get_tensor_buffer(output_tensor));
    int data_size = get_tensor_buffer_size(output_tensor) / sizeof(uint8_t);

    image img_out = imread(image_file);
    for (int i = 0; i < data_size / 2; i++)
    {
        int x = (int)(((float)data[2 * i] - (float)output_zp) * output_scale * (float)img_out.w / 144.f);
        int y = (int)(((float)data[2 * i + 1] - (float)output_zp) * output_scale * (float)img_out.h / 144.f);
        draw_circle(img_out, x, y, 2, 0, 255, 0);
    }

    save_image(img_out, "landmark_out_timvx");

    postrun_graph(graph);
    destroy_graph(graph);
    release_tengine();

    return 0;
}