caffe目标检测模型训练全过程（三）目标检测第一步

遍历整图查找蝴蝶位置

2018/04/22

训练模型对于识别背景和蝴蝶有比较好的效果，基本对不会识别错误，接下来，将通过整图遍历的原始而又野蛮的方式对一张原始图片进行处理，进而查找到蝴蝶的具体位置。具体思路如下图。对原图进行缩放成理想大小，例如，

最小边长缩放为227*6像素，最大边长等比缩放：

设置滑动窗大小为227*227像素

设置滑动窗移动步长为

# 自动step_size

scale =min(img.shape[0],(227*6.0)/img.shape[1])*1.0

step_size = int(min(img.shape[0]*scale,img.shape[1]*scale))/20

print("scale",img.shape[0]*scale,img.shape[1]*scale)

#resize image

scale_img = cv2.resize(img,((int(img.shape[1] * scale), int(img.shape[0] * scale))))

 

然后通过模型预测每个滑动框的类别，若是蝴蝶（类别1）的概率大于99%，则记录滑动框位置，其后绘制在图片上

具体代码：

# 自动step_size

        scale =min(img.shape[0],(227*6.0)/img.shape[1])*1.0

        print(scale)

        

        step_size = int(min(img.shape[0]*scale,img.shape[1]*scale))/20

        print("scale",img.shape[0]*scale,img.shape[1]*scale)

        #resize image

        scale_img = cv2.resize(img,((int(img.shape[1] * scale), int(img.shape[0] * scale))))

        temp2=scale_img[:]

        print(scale_img.shape,temp2.shape,img.shape)

        

        

        while x+227 < img.shape[0]*scale:

            while y+227 < img.shape[1]*scale:

                # cv2.rectangle(scale_img,(x,y),(x+227,y+227),(0,0,255),1)

                temp = scale_img[x:x+227,y:y+227]

                cv2.imwrite("scale_img.jpg",temp)

                im = caffe.io.load_image('scale_img.jpg')

                print(im.shape)

           

                transformer = caffe.io.Transformer({'data': net_full_conv.blobs['data'].data.shape})

                #transformer.set_mean('data', np.load(caffe_root + 'python/caffe/imagenet/ilsvrc_2012_mean.npy').mean(1).mean(1))

                transformer.set_transpose('data', (2,0,1))

                transformer.set_channel_swap('data', (2,1,0))

                transformer.set_raw_scale('data', 255.0)

                

                transformed_image = transformer.preprocess('data', im)

                net_full_conv.blobs['data'].data[...] = transformed_image

                net_full_conv.blobs['data'].reshape(1,3,227,227)

 

                ### perform classification

                out = net_full_conv.forward()

                # make classification map by forward and print prediction indices at each location

                # out = net_full_conv.forward_all(data=np.asarray([transformer.preprocess('data', im)]))

                print (out['prob'],(x,y))

                if (out['prob'][0][1]>0.99):

                    cv2.rectangle(temp2,(y,x),(y+227,x+227),(0,0,255),2)

                # print (out['prob'][0,1].shape,(x,y))

                del im

                y += step_size                

            y = 0

            x += step_size

        cv2.imwrite(imgFile.split('.')[0]+"_result.jpg",temp2)

其程序运行效果如下：