Faster RCNN 运行自己的数据,刚开始正常,后来就报错: Index exceeds matrix dimensions. Error in ori_demo (line 114) boxes_cell{i} = [boxes(:, (1+(i-1)*4):(i*4)), scores(

简介: function script_faster_rcnn_demo() close all; clc; clear mex; clear is_valid_handle; % to clear init_key run(fullfile(filepart...
function script_faster_rcnn_demo()
close all;       clc;      clear mex;          clear is_valid_handle; % to clear init_key
run(fullfile(fileparts(fileparts(mfilename('fullpath'))), 'startup'));
%% -------------------- CONFIG --------------------
opts.caffe_version          = 'caffe_faster_rcnn';
opts.gpu_id                 = auto_select_gpu;
active_caffe_mex(opts.gpu_id, opts.caffe_version);

opts.per_nms_topN           = 6000;
opts.nms_overlap_thres      = 0.7;
opts.after_nms_topN         = 300;
opts.use_gpu                = true;

opts.test_scales            = 600;

%% -------------------- INIT_MODEL --------------------
% model_dir                   = fullfile(pwd, 'output', 'faster_rcnn_final', 'faster_rcnn_VOC0712_vgg_16layers'); %% VGG-16
model_dir                   = fullfile(pwd, 'output', 'faster_rcnn_final', 'faster_rcnn_VOC0712_ZF'); %% ZF
proposal_detection_model    = load_proposal_detection_model(model_dir);

proposal_detection_model.conf_proposal.test_scales = opts.test_scales;
proposal_detection_model.conf_detection.test_scales = opts.test_scales;
if opts.use_gpu
    proposal_detection_model.conf_proposal.image_means = gpuArray(proposal_detection_model.conf_proposal.image_means);
    proposal_detection_model.conf_detection.image_means = gpuArray(proposal_detection_model.conf_detection.image_means);
end

% caffe.init_log(fullfile(pwd, 'caffe_log'));
% proposal net
rpn_net = caffe.Net(proposal_detection_model.proposal_net_def, 'test');
rpn_net.copy_from(proposal_detection_model.proposal_net);
% fast rcnn net
fast_rcnn_net = caffe.Net(proposal_detection_model.detection_net_def, 'test');
fast_rcnn_net.copy_from(proposal_detection_model.detection_net);

% set gpu/cpu
if opts.use_gpu
    caffe.set_mode_gpu();
else
    caffe.set_mode_cpu();
end       

%% -------------------- WARM UP --------------------
% the first run will be slower; use an empty image to warm up

for j = 1:2 % we warm up 2 times
    im = uint8(ones(375, 500, 3)*128);
    if opts.use_gpu
        im = gpuArray(im);
    end
    [boxes, scores]             = proposal_im_detect(proposal_detection_model.conf_proposal, rpn_net, im);
    aboxes                      = boxes_filter([boxes, scores], opts.per_nms_topN, opts.nms_overlap_thres, opts.after_nms_topN, opts.use_gpu);
    if proposal_detection_model.is_share_feature
        [boxes, scores]             = fast_rcnn_conv_feat_detect(proposal_detection_model.conf_detection, fast_rcnn_net, im, ...
            rpn_net.blobs(proposal_detection_model.last_shared_output_blob_name), ...
            aboxes(:, 1:4), opts.after_nms_topN);
    else
        [boxes, scores]             = fast_rcnn_im_detect(proposal_detection_model.conf_detection, fast_rcnn_net, im, ...
            aboxes(:, 1:4), opts.after_nms_topN);
    end
end

%% -------------------- TESTING --------------------
% im_names = {'001763.jpg', '004545.jpg', '000542.jpg', '000456.jpg', '001150.jpg'};
% these images can be downloaded with fetch_faster_rcnn_final_model.m
image_path = '/media/wangxiao/Elements/image_segment_backup/ImagesData223/Pedestrian/172.19.199.223/';
file1 = dir(image_path);

for ii = 3:size(file1, 1)
    new_path1 = [image_path, file1(ii).name, '/'];
    file2 = dir(new_path1);
    for jj = 3:size(file2, 1)
        new_path2 = [new_path1, file2(jj).name, '/'];
        file3 = dir(new_path2);
        for kk = 3:size(file3, 1)
            im = imread([new_path2, file3(kk).name]);
            running_time = [];
            % for j = 1:length(im_names)
            % im = imread(fullfile(pwd, im_names{j}));
    
            if opts.use_gpu
                 im = gpuArray(im);
            end
    
    % test proposal
    th = tic();
    [boxes, scores]             = proposal_im_detect(proposal_detection_model.conf_proposal, rpn_net, im);
    t_proposal = toc(th);
    th = tic();
    aboxes                      = boxes_filter([boxes, scores], opts.per_nms_topN, opts.nms_overlap_thres, opts.after_nms_topN, opts.use_gpu);
    t_nms = toc(th);
    
    % test detection
    th = tic();
    if proposal_detection_model.is_share_feature
        [boxes, scores]             = fast_rcnn_conv_feat_detect(proposal_detection_model.conf_detection, fast_rcnn_net, im, ...
            rpn_net.blobs(proposal_detection_model.last_shared_output_blob_name), ...
            aboxes(:, 1:4), opts.after_nms_topN);
    else
        [boxes, scores]             = fast_rcnn_im_detect(proposal_detection_model.conf_detection, fast_rcnn_net, im, ...
            aboxes(:, 1:4), opts.after_nms_topN);
    end
    t_detection = toc(th);
    
%     fprintf('%s (%dx%d): time %.3fs (resize+conv+proposal: %.3fs, nms+regionwise: %.3fs)\n', im_names{j}, ...
%         size(im, 2), size(im, 1), t_proposal + t_nms + t_detection, t_proposal, t_nms+t_detection);
%     running_time(end+1) = t_proposal + t_nms + t_detection;
    
    % visualize
    classes = proposal_detection_model.classes;
    boxes_cell = cell(length(classes), 1);
    thres = 0.6;
    for i = 1:length(boxes_cell)
        boxes_cell{i} = [boxes(:, (1+(i-1)*4):(i*4)), scores(:, i)];
        boxes_cell{i} = boxes_cell{i}(nms(boxes_cell{i}, 0.3), :);
        
        I = boxes_cell{i}(:, 5) >= thres;
        boxes_cell{i} = boxes_cell{i}(I, :);
    end
    figure(j);
    
    
    
    [location, label, score] = output(im, boxes_cell, classes, 'voc');
    if (score==0)
       continue;
    else
       disp(file3(kk).name);
       save(['./mat results/' file3(kk).name '.mat' ], 'location', 'label', 'score', 'im');
    end
    
    showboxes(im, boxes_cell, classes, 'voc');
    pause(0.1);
        end
    end
end


% fprintf('mean time: %.3fs\n', mean(running_time));

caffe.reset_all(); 
clear mex;

end

-------------------------------------------------------
---------------------------------------------------------

function proposal_detection_model = load_proposal_detection_model(model_dir)
    ld                          = load(fullfile(model_dir, 'model'));
    proposal_detection_model    = ld.proposal_detection_model;
    clear ld;
    
    proposal_detection_model.proposal_net_def ...
                                = fullfile(model_dir, proposal_detection_model.proposal_net_def);
    proposal_detection_model.proposal_net ...
                                = fullfile(model_dir, proposal_detection_model.proposal_net);
    proposal_detection_model.detection_net_def ...
                                = fullfile(model_dir, proposal_detection_model.detection_net_def);
    proposal_detection_model.detection_net ...
                                = fullfile(model_dir, proposal_detection_model.detection_net);
    
end

function aboxes = boxes_filter(aboxes, per_nms_topN, nms_overlap_thres, after_nms_topN, use_gpu)
    % to speed up nms
    if per_nms_topN > 0
        aboxes = aboxes(1:min(length(aboxes), per_nms_topN), :);
    end
    % do nms
    if nms_overlap_thres > 0 && nms_overlap_thres < 1
        aboxes = aboxes(nms(aboxes, nms_overlap_thres, use_gpu), :);       
    end
    if after_nms_topN > 0
        aboxes = aboxes(1:min(length(aboxes), after_nms_topN), :);
    end
end

 

matlab命令窗口,显示: 刚开始都是正常的,如下:

fast_rcnn startup done
GPU 1: free memory 3824902144
Use GPU 1
Warning: Specified caffe folder (/home/wangxiao/Downloads/faster_rcnn-master/experiments/external/caffe/matlab/caffe_faster_rcnn)
is not exist, change to default one (/home/wangxiao/Downloads/faster_rcnn-master/experiments/external/caffe/matlab)
> In active_caffe_mex at 19
  In ori_demo at 7
20150301095338.jpg
20150301095445.jpg
20150301095535.jpg
20150301095543.jpg
20150301095603.jpg
20150301095613.jpg
20150301095617.jpg
20150301095632.jpg
20150301095635.jpg
20150301095646.jpg
20150301095656.jpg
20150301095659.jpg
20150301095711.jpg
20150301095714.jpg
20150301095717.jpg
20150301095720.jpg
20150301095723.jpg
20150301095726.jpg
20150301095729.jpg
20150301095734.jpg
20150301095741.jpg
20150301095750.jpg


Index exceeds matrix dimensions.

Error in ori_demo (line 114)
        boxes_cell{i} = [boxes(:, (1+(i-1)*4):(i*4)), scores(:, i)];
 

但是,跑着跑着就出问题了,索引超过矩阵维度是什么鬼??只好硬着头皮一句一句的调试了,fuck 。。。。

 

但是,搞不定啊。。只知道了一个:非极大值抑制(NMS) 

为什么会出错呢??? 之前跑的好好地啊。。。提示超过矩阵范围的那一句是作者自己写的。。。怎么会错呢?怎么会错?怎么会?怎么?怎? ?

 

 

 不搞了,先把原始的图像,整理成一个文件夹在做处理:

clc; close all;

path = '/media/wangxiao/Elements/image_segment_backup/';
savePath = '/media/wangxiao/Elements/wang xiao/additional_data/';
camera = dir(path);

txt_path = '/media/wangxiao/Elements/wang xiao/';
txt_file = fopen([txt_path, 'log_file.txt'], 'a');

for i = 3:length(camera)
    disp(['camera ', num2str(i-2), '---']);
    fprintf(txt_file, '%s \n ', num2str(i-2));
    path2 = [path, camera(i).name, '/'];
    file1 = dir(path2);
    for ii = 5:size(file1, 1)
        disp(['    ',  file1(ii).name, '---']);
        fprintf(txt_file, '%s  \n', file1(ii).name);
        new_path = [path2, file1(ii).name, '/'];
        file2 = dir(new_path);
        for j = 3:size(file2, 1)
            disp(['    ',  file2(j).name, '---']);
            fprintf(txt_file, '%s  \n', file2(j).name);
            new_path2 = [new_path, file2(j).name, '/'];
            file3 = dir(new_path2);
            for k = 3:size(file3, 1)
                disp(['    ',  file3(k).name, '---']);
                fprintf(txt_file, '%s  \n', file3(k).name);
                new_path3 = [new_path2, file3(k).name , '/'];
                file4 = dir(new_path3);
                for r = 3:size(file4, 1)
                    disp(['    ',  file4(r).name, '---']);
                     fprintf(txt_file, '%s  \n', file4(r).name);
                    new_path4 = [new_path3, file4(r).name, '/'];
                    file5 = dir(new_path4);
                    for w = 3:size(file5, 1)
                        if (imread([new_path4, file5(w).name]))
                            im = imread([new_path4, file5(w).name]);
                            imshow(im);
                            imwrite(im, [savePath, file5(w).name]);
                        else
                            continue;
                            
                        end
                    end
                end
            end
        end
    end
end

fclose(txt_file);

 

 

 找到原因了,妈的,原来是因为,输入图像的大小不一致导致的,奇怪了,只要加一句: im = imresize(im, [127 127]); 将输入的图像统一resize成 固定的大小,即可。。。简单 粗暴 但是,不解其惑 。。。。

 

 

...

相关实践学习
部署Stable Diffusion玩转AI绘画(GPU云服务器)
本实验通过在ECS上从零开始部署Stable Diffusion来进行AI绘画创作,开启AIGC盲盒。
相关文章
|
6月前
|
机器学习/深度学习 算法 安全
FRCNN来袭 | Faster RCNN与FCN永不遗忘,联邦学习+边缘数据既保护隐私也提升性能
FRCNN来袭 | Faster RCNN与FCN永不遗忘,联邦学习+边缘数据既保护隐私也提升性能
272 0
|
6月前
|
机器学习/深度学习 TensorFlow 算法框架/工具
PYTHON TENSORFLOW 2二维卷积神经网络CNN对图像物体识别混淆矩阵评估|数据分享
PYTHON TENSORFLOW 2二维卷积神经网络CNN对图像物体识别混淆矩阵评估|数据分享
|
6月前
|
机器学习/深度学习 存储 监控
数据分享|Python卷积神经网络CNN身份识别图像处理在疫情防控下口罩识别、人脸识别
数据分享|Python卷积神经网络CNN身份识别图像处理在疫情防控下口罩识别、人脸识别
|
19天前
|
机器学习/深度学习 自然语言处理 前端开发
前端神经网络入门:Brain.js - 详细介绍和对比不同的实现 - CNN、RNN、DNN、FFNN -无需准备环境打开浏览器即可测试运行-支持WebGPU加速
本文介绍了如何使用 JavaScript 神经网络库 **Brain.js** 实现不同类型的神经网络,包括前馈神经网络(FFNN)、深度神经网络(DNN)和循环神经网络(RNN)。通过简单的示例和代码,帮助前端开发者快速入门并理解神经网络的基本概念。文章还对比了各类神经网络的特点和适用场景,并简要介绍了卷积神经网络(CNN)的替代方案。
|
4月前
|
机器学习/深度学习 PyTorch 算法框架/工具
图神经网络是一类用于处理图结构数据的神经网络。与传统的深度学习模型(如卷积神经网络CNN和循环神经网络RNN)不同,
图神经网络是一类用于处理图结构数据的神经网络。与传统的深度学习模型(如卷积神经网络CNN和循环神经网络RNN)不同,
|
5月前
|
机器学习/深度学习 算法 计算机视觉
基于CNN卷积神经网络的金融数据预测matlab仿真,带GUI界面,对比BP,RBF,LSTM
这是一个基于MATLAB2022A的金融数据预测仿真项目,采用GUI界面,比较了CNN、BP、RBF和LSTM四种模型。CNN和LSTM作为深度学习技术,擅长序列数据预测,其中LSTM能有效处理长序列。BP网络通过多层非线性变换处理非线性关系,而RBF网络利用径向基函数进行函数拟合和分类。项目展示了不同模型在金融预测领域的应用和优势。
|
6月前
|
机器学习/深度学习 算法 TensorFlow
【视频】神经网络正则化方法防过拟合和R语言CNN分类手写数字图像数据MNIST|数据分享
【视频】神经网络正则化方法防过拟合和R语言CNN分类手写数字图像数据MNIST|数据分享
|
6月前
|
机器学习/深度学习 数据采集 TensorFlow
R语言KERAS深度学习CNN卷积神经网络分类识别手写数字图像数据(MNIST)
R语言KERAS深度学习CNN卷积神经网络分类识别手写数字图像数据(MNIST)
|
6月前
|
机器学习/深度学习 数据可视化 定位技术
PrObeD方法开源 | 主动方法助力YOLOv5/Faster RCNN/DETR在COCO/GOD涨点
PrObeD方法开源 | 主动方法助力YOLOv5/Faster RCNN/DETR在COCO/GOD涨点
80 0
|
6月前
|
机器学习/深度学习 监控 算法
【Keras计算机视觉】Faster R-CNN神经网络实现目标检测实战(附源码和数据集 超详细)
【Keras计算机视觉】Faster R-CNN神经网络实现目标检测实战(附源码和数据集 超详细)
100 0

热门文章

最新文章

下一篇
无影云桌面