make_file.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import cv2\n",
    "import json\n",
    "import os\n",
    "import  numpy as np\n",
    "import time\n",
    "import glob\n",
    "#########注意input json路径  \n",
    "result = \"/home/admin/jupyter/Demo/Models/detectron2_bai/outs/out_cascade_mask_rcnn_X_152/inference/my_dataset_test.json\"\n",
    "#\"/home/admin/jupyter/Demo/Models/detectron2_bai/outs/out_cascade_mask_rcnn_X_152/inference/my_dataset_test.json\"  #input_json\n",
    "result1 = result # \"/home/admin/jupyter/Demo/Models/detectron2_bai/outs/out_cascade_mask_rcnn_X_152_cl4/inference/my_dataset_test.json\"  #input_json\n",
    "result2 = result #\"/home/admin/jupyter/Demo/Models/detectron2_bai/outs/out_cascade_mask_rcnn_R_50_CA/inference/my_dataset_test.json\"  #input_json\n",
    "########测试集json###\n",
    "refjson=  '/home/admin/jupyter/Demo/DataSets/Data/test.json' #\"/home/admin/jupyter/Demo/DataSets/Data/test_ts.json\"\n",
    "refdir=  r'/home/admin/jupyter/Demo/DataSets/Data/images_test_json/'#/home/admin/jupyter/Demo/DataSets/Data/images_test_ts_json/' #' \n",
    "####################"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "################## load results & build category dict###########\n",
    "start=time.time()\n",
    "with  open(result, \"r\") as f:\n",
    "    results = json.load(f)\n",
    "print('###Results e.g. %s'% results[:4])\n",
    "print('!!!Total: %d'%len(results),' cost time s: ',time.time()-start)\n",
    "#####加\n",
    "with  open(result1, \"r\") as f:\n",
    "    results1 = json.load(f)\n",
    "with  open(result2, \"r\") as f:\n",
    "    results2 = json.load(f)\n",
    "\n",
    "start=time.time()\n",
    "with  open(refjson, \"r\") as f:\n",
    "    j=json.load(f)\n",
    "    print('###Test annotations: ',j['annotations'][:4])\n",
    "    categories_list=j['categories']\n",
    "    im_list = j['images']\n",
    "#print('###Test images e.g. %s'%im_list[:4],'\\tTotal: %d'%len(im_list),'cost time s:',time.time()-start)\n",
    "categories_dict={}\n",
    "for c in categories_list:\n",
    "    categories_dict[c['id']] = c['name'] #建立字典 id对应 类别\n",
    "print('Categories: %s'%categories_dict)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#建立id-name字典\n",
    "images_dict = {}\n",
    "start=time.time()\n",
    "n=0\n",
    "for im_str in im_list:\n",
    "    name=im_str['file_name'].split('.')[0]\n",
    "    with  open(refdir + name + '.json', \"r\") as f:\n",
    "            js = json.load(f)\n",
    "#             print(js)\n",
    "    if not js.__contains__('xy'): \n",
    "        images_dict[im_str['id']] = [js['at_20_Original_image'][0],js['at_20_Original_image'][1],name.split('_')[0]]\n",
    "    else:\n",
    "        images_dict[im_str['id']] = [js['xy'][0],js['xy'][1],js['Original_name']] #建立字典 id 对应 原图x,y 原图名字\n",
    "\n",
    "print('建立字典,共%d个,e.g. %s'%(len(images_dict),images_dict[1]),'cost time s:',time.time()-start) #70s\n",
    "print(im_list[:5])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import copy\n",
    "def chuli5 (bboxes):  ###chuli 2与 4结合\n",
    "    \n",
    "    tboxes =  bboxes  \n",
    "    for i ,box in enumerate(bboxes):\n",
    "        if bboxes[i][\"p\"] == 0.0:\n",
    "            continue        \n",
    "        for j,tbox in enumerate(tboxes):\n",
    "            if j==i:\n",
    "                continue\n",
    "            if bboxes[j][\"p\"] == 0.0: # \n",
    "                continue\n",
    "            if box[\"class\"]!=tbox[\"class\"]:   ##加 对不同类 分离处理\n",
    "                continue\n",
    "            ##l两筐没关系\n",
    "            if box[\"x\"]+box['w']<=tbox[\"x\"] or box[\"y\"]+box['h']<=tbox[\"y\"] or box[\"x\"]>=tbox[\"x\"]+tbox[\"w\"] or box[\"y\"]>=tbox[\"y\"]+tbox[\"h\"]: \n",
    "                continue\n",
    "            ##在里面   念珠菌先处理 包含关系 和后面有点不一样\n",
    "#             if box[\"class\"]== \"Candida\" and box[\"x\"]>=tbox[\"x\"] and box[\"y\"]>=tbox[\"y\"] and box[\"x\"] + box[\"w\"]<=tbox[\"x\"]+tbox[\"w\"] and box[\"y\"] + box[\"h\"]<=tbox[\"y\"]+tbox[\"h\"] :\n",
    "#                 if box[\"p\"]>=tbox[\"p\"]+0.2 :  ###  +0.2\n",
    "#                     bboxes[j][\"p\"] = 0\n",
    "#                 else :\n",
    "#                     bboxes[i][\"p\"] = 0\n",
    "#                 continue\n",
    "            ####得到 iou   公式改进\n",
    "            in_h = min(box['y']+box['h'],tbox['y']+tbox['h'])-max(box['y'],tbox['y'])\n",
    "            in_w = min(box['x']+box['w'],tbox['x']+tbox['w'])-max(box['x'],tbox['x'])\n",
    "            union = box['w']*box['h']+tbox['w']*tbox['h']-in_h*in_w\n",
    "            iou = in_h*in_w/union\n",
    "            # 单独对 念珠菌处理 更容易放大操作\n",
    "            if box[\"class\"]== \"Candida\":   # 单独对 念珠菌处理 更容易放大操作\n",
    "                ##在里面\n",
    "                if box[\"x\"]>=tbox[\"x\"] and box[\"y\"]>=tbox[\"y\"] and box[\"x\"] + box[\"w\"]<=tbox[\"x\"]+tbox[\"w\"] and box[\"y\"] + box[\"h\"]<=tbox[\"y\"]+tbox[\"h\"] :\n",
    "                    if box[\"p\"]>=tbox[\"p\"]+0.2 :  ###  +0.2\n",
    "                        bboxes[j][\"p\"] = 0\n",
    "                    else :\n",
    "                        bboxes[i][\"p\"] = 0\n",
    "                    continue\n",
    "                if iou >= 0.95 :  ###0.8         #iou 高                                    \n",
    "                    if box[\"p\"]>=tbox[\"p\"]:  \n",
    "                        bboxes[j][\"p\"] = 0\n",
    "                    else :\n",
    "                        bboxes[i][\"p\"] = 0\n",
    "                    continue\n",
    "                elif iou >0.4 and iou < 0.95 and box[\"class\"]==tbox[\"class\"] and box[\"p\"]>0.3 and tbox[\"p\"]>0.3: #iou中等，且同类，且置信度高\n",
    "                    bboxes[i][\"x\"] = min(box[\"x\"],tbox[\"x\"])\n",
    "                    bboxes[i][\"y\"] = min(box[\"y\"],tbox[\"y\"])\n",
    "                    bboxes[i][\"w\"] = abs(box[\"x\"]-tbox[\"x\"])+tbox[\"w\"]\n",
    "                    bboxes[i][\"h\"] = abs(box[\"y\"]-tbox[\"y\"])+tbox[\"h\"]\n",
    "                    bboxes[i][\"p\"] = max(box[\"p\"],tbox[\"p\"])\n",
    "                    bboxes[j][\"p\"] = 0\n",
    "                    continue\n",
    "                elif iou >0.4 and iou <0.95 :   #iou中等 且同类，置信度不高\n",
    "                    if box[\"p\"]>=tbox[\"p\"]:  \n",
    "                        bboxes[j][\"p\"] = 0\n",
    "                    else :\n",
    "                        bboxes[i][\"p\"] = 0\n",
    "                    continue\n",
    "                else:\n",
    "                    pass\n",
    "\n",
    "                ## 对大小框处理 大框包括小框太多\n",
    "                s1 = box[\"w\"] *box[\"h\"]\n",
    "                s2 = tbox[\"w\"] *tbox[\"h\"]\n",
    "                ss = in_h*in_w\n",
    "\n",
    "                if (ss/s1 >0.6 )and ( s2/s1>3) : ##s1 太小  0.5 改0.8   \n",
    "                    if tbox[\"p\"]> 0.5:\n",
    "                        bboxes[i][\"p\"] = 0\n",
    "                    elif box[\"p\"]>=tbox[\"p\"]+0.2 :  ###  +0.2\n",
    "                        bboxes[j][\"p\"] = 0\n",
    "                    else :\n",
    "                        bboxes[i][\"p\"] = 0\n",
    "                    continue\n",
    "                elif (ss/s2 >0.6 )and ( s1/s2>3) :  ##s2 太小\n",
    "                    if box[\"p\"]> 0.5:\n",
    "                        bboxes[j][\"p\"] = 0\n",
    "                    elif box[\"p\"]+0.2<=tbox[\"p\"] :  ###  +0.2\n",
    "                        bboxes[i][\"p\"] = 0\n",
    "                    else :\n",
    "                        bboxes[j][\"p\"] = 0\n",
    "                    continue\n",
    "                else:\n",
    "                    pass\n",
    "            ############  对其他类的处理\n",
    "            if iou >= 0.8 :  ###0.8         #iou 高                                    \n",
    "                if box[\"p\"]>=tbox[\"p\"]:  \n",
    "                    bboxes[j][\"p\"] = 0\n",
    "                else :\n",
    "                    bboxes[i][\"p\"] = 0\n",
    "                continue\n",
    "            elif iou >0.4 and iou <0.8 and box[\"p\"]>0.4 and tbox[\"p\"]>0.4: #iou中等，且同类，且置信度高\n",
    "                bboxes[i][\"x\"] = min(box[\"x\"],tbox[\"x\"])\n",
    "                bboxes[i][\"y\"] = min(box[\"y\"],tbox[\"y\"])\n",
    "                bboxes[i][\"w\"] = abs(box[\"x\"]-tbox[\"x\"])+tbox[\"w\"]\n",
    "                bboxes[i][\"h\"] = abs(box[\"y\"]-tbox[\"y\"])+tbox[\"h\"]\n",
    "                bboxes[i][\"p\"] = max(box[\"p\"],tbox[\"p\"])\n",
    "                bboxes[j][\"p\"] = 0\n",
    "                continue\n",
    "            elif iou >0.4 and iou <0.8 :   #iou中等 且同类，置信度不高\n",
    "                if box[\"p\"]>=tbox[\"p\"]:  \n",
    "                    bboxes[j][\"p\"] = 0\n",
    "                else :\n",
    "                    bboxes[i][\"p\"] = 0\n",
    "                continue\n",
    "            else:\n",
    "                pass\n",
    "            \n",
    "    for i ,box in enumerate(bboxes):\n",
    "        if bboxes[i][\"p\"] == 0.0:\n",
    "            continue\n",
    "        if box[\"class\"]== \"Candida\":  ##上面全面处理了\n",
    "            continue\n",
    "        for j,tbox in enumerate(tboxes):\n",
    "            if j==i:\n",
    "                continue\n",
    "            if bboxes[j][\"p\"] == 0.0: # \n",
    "                continue\n",
    "            if box[\"class\"]!=tbox[\"class\"]:   ##加 对不同类 分离处理\n",
    "                continue\n",
    "            ##l两筐没关系\n",
    "            if box[\"x\"]+box['w']<=tbox[\"x\"] or box[\"y\"]+box['h']<=tbox[\"y\"] or box[\"x\"]>=tbox[\"x\"]+tbox[\"w\"] or box[\"y\"]>=tbox[\"y\"]+tbox[\"h\"]: \n",
    "                continue\n",
    "            ##在里面 \n",
    "            if  box[\"x\"]>=tbox[\"x\"] and box[\"y\"]>=tbox[\"y\"] and box[\"x\"] + box[\"w\"]<=tbox[\"x\"]+tbox[\"w\"] and box[\"y\"] + box[\"h\"]<=tbox[\"y\"]+tbox[\"h\"] :\n",
    "                if tbox[\"p\"]> 0.5:\n",
    "                    bboxes[i][\"p\"] = 0\n",
    "                elif box[\"p\"]>=tbox[\"p\"]+0.2 :  ###  +0.2\n",
    "                    bboxes[j][\"p\"] = 0\n",
    "                else :\n",
    "                    bboxes[i][\"p\"] = 0\n",
    "                continue\n",
    "            ## 对大小框处理 大框包括小框太多\n",
    "            s1 = box[\"w\"] *box[\"h\"]\n",
    "            s2 = tbox[\"w\"] *tbox[\"h\"]\n",
    "            in_h = min(box['y']+box['h'],tbox['y']+tbox['h'])-max(box['y'],tbox['y'])\n",
    "            in_w = min(box['x']+box['w'],tbox['x']+tbox['w'])-max(box['x'],tbox['x'])\n",
    "            ss = in_h * in_w\n",
    "            \n",
    "            if (ss/s1 >0.8 )and ( s2/s1>4) : ##s1 太小  0.5 改0.8   \n",
    "                if tbox[\"p\"]> 0.5:\n",
    "                    bboxes[i][\"p\"] = 0\n",
    "                elif box[\"p\"]>=tbox[\"p\"]+0.2 :  ###  +0.2\n",
    "                    bboxes[j][\"p\"] = 0\n",
    "                else :\n",
    "                    bboxes[i][\"p\"] = 0\n",
    "                continue\n",
    "            elif (ss/s2 >0.8 )and ( s1/s2>4) :  ##s2 j 小  i 大\n",
    "                if box[\"p\"]> 0.5:\n",
    "                    bboxes[j][\"p\"] = 0\n",
    "                elif box[\"p\"]+0.2<=tbox[\"p\"] :  ###  +0.2\n",
    "                    bboxes[i][\"p\"] = 0\n",
    "                else :\n",
    "                    bboxes[j][\"p\"] = 0\n",
    "                continue\n",
    "            else:\n",
    "                pass\n",
    "    return bboxes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def arrays2dicts(boxes,thres=0):\n",
    "    \"\"\"\n",
    "    boxes:a dict of arrays, key as class and each box array:x,y,w,h,p\n",
    "    return:a list of dicts,and each box dict:x,y,w,h,p,class \n",
    "    \"\"\"\n",
    "    out = []\n",
    "    categories={1: 'ASC-H', 2: 'ASC-US', 3: 'HSIL', 4: 'LSIL', 5: 'Candida', 6: 'Trichomonas'}\n",
    "    counts={'HSIL':0, 'LSIL': 0, 'ASC-H': 0, 'ASC-US': 0,'Candida':0, 'Trichomonas':0}\n",
    "    if not isinstance(boxes,dict):\n",
    "        boxes={'all':boxes}\n",
    "    for c,boxes_ in boxes.items():\n",
    "        if isinstance(thres,dict):\n",
    "            threshold=thres[c]\n",
    "        elif isinstance(thres,float):\n",
    "#             assert(isinstance(thres,float),'should be dict or float!')\n",
    "            threshold=thres\n",
    "        else:\n",
    "            raise Exception('thres should be dict or float!')\n",
    "        for i in range(len(boxes_)):\n",
    "            b = boxes_[i]\n",
    "            # b x y w h p\n",
    "            if b[4]<=threshold:\n",
    "                continue\n",
    "            box = {}\n",
    "            cls=categories[b[5]]\n",
    "            box['x'] = b[0]\n",
    "            box['y'] = b[1]\n",
    "            box['w'] = b[2] - box['x']\n",
    "            box['h'] = b[3] - box['y']\n",
    "            box['p'] = b[4]\n",
    "            box['class'] =cls\n",
    "            out.append(box)\n",
    "            counts[cls]+=1\n",
    "            \n",
    "    return out,counts\n",
    "\n",
    "\n",
    "def dicts2arrays(boxes):\n",
    "    \"\"\"\n",
    "    boxes:a list of dicts,and each box dict:x,y,w,h,p,class\n",
    "    return: a dict of arrays, let class as key and each box array:x,y,w,h,p,class\n",
    "    \"\"\"\n",
    "    out = {}\n",
    "    categories={'ASC-H':1, 'ASC-US':2, 'HSIL':3, 'LSIL':4, 'Candida':5, 'Trichomonas':6}\n",
    "    for box in boxes:\n",
    "        # box x y w h p class\n",
    "        b = np.zeros(6)\n",
    "        b[0] = box['x']\n",
    "        b[1] = box['y']\n",
    "        b[2] = box['x'] + box['w']\n",
    "        b[3] = box['y'] + box['h']\n",
    "        b[4] = box['p']\n",
    "        b[5] = categories[box['class']]\n",
    "        b = np.expand_dims(b, axis=0)\n",
    "        if box['class'] not in out:\n",
    "            out[box['class']]=b\n",
    "        else:\n",
    "            out[box['class']] = np.concatenate((out[box['class']], b), axis=0)\n",
    "    return out\n",
    "\n",
    "\n",
    "def nms(boxes, threshold=0.5, method='original', sigma=0.5):\n",
    "    #boxes :an array of x,y,w,h,p\n",
    "    #     print(boxes.shape)\n",
    "    # coordinates of bounding boxes\n",
    "    if (len(boxes))==0:\n",
    "        return boxes\n",
    "    s_x = boxes[:, 0]\n",
    "    s_y = boxes[:, 1]\n",
    "    e_x = boxes[:, 2]\n",
    "    e_y = boxes[:, 3]\n",
    "    # Confidence scores of bounding boxes\n",
    "    score = boxes[:, 4]\n",
    "    # Picked bounding boxes\n",
    "    #     picked_boxes = []\n",
    "    # Compute areas of bounding boxes\n",
    "    areas = (e_x - s_x + 1) * (e_y - s_y + 1)\n",
    "    # Sort by confidence score of bounding boxes\n",
    "    order = np.argsort(score)\n",
    "\n",
    "    # Iterate bounding boxes\n",
    "    while order.size > 0:\n",
    "        # The index of largest confidence score\n",
    "        index = order[-1]\n",
    "\n",
    "        # Pick the bounding box with largest confidence score\n",
    "        #         picked_boxes.append(bboxes[index])\n",
    "\n",
    "        # Compute ordinates of intersection-over-union(IOU)\n",
    "        x1 = np.maximum(s_x[index], s_x[order[:-1]])\n",
    "        x2 = np.minimum(e_x[index], e_x[order[:-1]])\n",
    "        y1 = np.maximum(s_y[index], s_y[order[:-1]])\n",
    "        y2 = np.minimum(e_y[index], e_y[order[:-1]])\n",
    "\n",
    "        # Compute areas of intersection-over-union\n",
    "        w = np.maximum(0.0, x2 - x1 + 1)\n",
    "        h = np.maximum(0.0, y2 - y1 + 1)\n",
    "        intersection = w * h\n",
    "\n",
    "        # Compute the ratio between intersection and union\n",
    "        ratio = intersection / (areas[index] + areas[order[:-1]] - intersection)\n",
    "        if method == 'gaussian':\n",
    "            weight = np.exp(-(ratio * ratio) / sigma)\n",
    "        elif method == 'linear':\n",
    "            weight = 1 - ratio\n",
    "        elif method == 'original':\n",
    "            weight = np.ones_like(ratio)\n",
    "            weight[ratio > threshold] = 0\n",
    "        else:\n",
    "            weight = np.ones_like(ratio)\n",
    "        weight[ratio > 0.7] = 0\n",
    "#         print(len(ratio),len(weight),len(order))\n",
    "        boxes[order[:-1], 4] *= weight\n",
    "        left = np.where(ratio < threshold)\n",
    "        order = order[left]\n",
    "    boxes=boxes[np.argsort(boxes[:, 4]),:]\n",
    "    return boxes\n",
    "def nms_by_class(bboxes, threshold=0.5, method='original', sigma=0.5,score_thres=0):\n",
    "    \"\"\"\n",
    "    bboxes: a list of boxes,and each box :x,y,w,h,p,class\n",
    "    \"\"\"\n",
    "    counts={'HSIL':0, 'LSIL': 0, 'ASC-H': 0, 'ASC-US': 0,'Candida':0, 'Trichomonas':0}\n",
    "    if len(bboxes) == 0:\n",
    "        return [],counts\n",
    "    # Bounding boxes\n",
    "    cboxes = dicts2arrays(bboxes)\n",
    "    oboxes={}\n",
    "    total_nonzero=0\n",
    "    log=''\n",
    "    for c,boxes in cboxes.items():\n",
    "        oboxes[c]=nms(boxes, threshold=threshold, method=method, sigma=sigma)\n",
    "        log+='%s:(non_zero:%d,max:%3f) ; '%(c,np.count_nonzero(oboxes[c][:,-2]),oboxes[c][-1][-2])\n",
    "    out,counts=arrays2dicts(oboxes,score_thres)\n",
    "    print('%s\\nFiltered %s'%(log,counts))\n",
    "    return out,counts\n",
    "def nms_all(bboxes, threshold=0.5, method='original', sigma=0.5,score_thres=0):\n",
    "    \"\"\"\n",
    "    bboxes: a list of boxes,and each box :x,y,w,h,p,class\n",
    "    \"\"\"\n",
    "    counts={'HSIL':0, 'LSIL': 0, 'ASC-H': 0, 'ASC-US': 0,'Candida':0, 'Trichomonas':0}\n",
    "    if len(bboxes) == 0:\n",
    "        return [],counts\n",
    "    # Bounding boxes\n",
    "    cboxes = dicts2arrays(bboxes)\n",
    "    total_nonzero=0\n",
    "    for i,(_,boxes) in enumerate(cboxes.items()):\n",
    "        if i==0:\n",
    "            oboxes=boxes\n",
    "        else:\n",
    "            oboxes=np.concatenate((oboxes, boxes), axis=0)\n",
    "    \n",
    "    oboxes=nms(oboxes, threshold=threshold, method=method, sigma=sigma)\n",
    "    out,counts=arrays2dicts(oboxes,score_thres)\n",
    "    print('Filtered %s'%(counts))\n",
    "    return out,counts"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# nms & merge \n",
    "outpath='./out_jsons/'#'./out_val_20_jsons/' './out_jsons/'\n",
    "if not os.path.exists(outpath):\n",
    "    os.makedirs(outpath)\n",
    "start=time.time()\n",
    "count=0\n",
    "total=0\n",
    "total_by_cls={'HSIL':0, 'LSIL': 0, 'ASC-H': 0, 'ASC-US': 0,'Candida':0, 'Trichomonas':0}\n",
    "merge=False#####\n",
    "sub_path1='./d191223/out_jsons/'\n",
    "# score_thres={'all':0.2,'HSIL':0.5, 'LSIL': 0.5, 'ASC-H': 0.3, 'ASC-US': 0.3,'Candida':0.2, 'Trichomonas':0.2} # 1213_cls_4\n",
    "# score_thres={'all':0.2,'HSIL':0.6, 'LSIL': 0.2, 'ASC-H': 0.6, 'ASC-US': 0.2,'Candida':0.2, 'Trichomonas':0.2} #1215_cls_2\n",
    "# score_thres={'all':0.2,'HSIL':0.6, 'LSIL': 0.2, 'ASC-H': 0.6, 'ASC-US': 0.6,'Candida':0.2, 'Trichomonas':0.2}#1217_cls_3\n",
    "# score_thres={'all':0.0,'HSIL':0.3, 'LSIL': 0.3, 'ASC-H': 0.3, 'ASC-US': 0.3,'Candida':0.3, 'Trichomonas':0.3}#1217_cls_6 建议 'Candida':0.5 考虑 'Trichomonas':0.2~0.3\n",
    "# score_thres={'all':0.0,'HSIL':0.3, 'LSIL': 0.3, 'ASC-H': 0.2, 'ASC-US': 0.2,'Candida':0.2, 'Trichomonas':0.2} #1224_cls_4\n",
    "score_thres={'all':0.0,'HSIL':0.5, 'LSIL': 0.3, 'ASC-H': 0.4, 'ASC-US': 0.4,'Candida':0.5, 'Trichomonas':0.2}\n",
    "for name,bboxes in out.items():\n",
    "    count+=1\n",
    "    print('No.%d %s.json total:%d'%(count,name,len(bboxes)))\n",
    "    print('nms_by_cls_1')\n",
    "    boxes,count_by_cls=nms_by_class(bboxes,0.4,method='linear',score_thres=score_thres) #  'gaussian'   ''   'original'   'linear'\n",
    "    num=len(boxes)\n",
    "    total_by_cls=dict(Counter(total_by_cls)+Counter(count_by_cls))\n",
    "    print('After nms_1 %s.json->total:%d'%(name,num))\n",
    "    #####mergerd with subjsons#################\n",
    "    if merge:\n",
    "        sub_jsons=os.path.join(sub_path1,name+'.json')\n",
    "        if os.path.exists(sub_jsons):\n",
    "            with open(sub_jsons,'r') as f:\n",
    "                subs=json.load(f)\n",
    "                #只取滴虫和念珠菌\n",
    "                subs,count_by_cls=nms_by_class(subs,0.4,method='',score_thres={'all':0.0,'HSIL':1, 'LSIL': 1, 'ASC-H': 1, 'ASC-US': 1,'Candida':0, 'Trichomonas':0} )\n",
    "                boxes+=subs\n",
    "                print('merged with %d ,get total:%d'%(len(subs),len(boxes)))\n",
    "                print('nms_by_cls_2')\n",
    "#                 boxes,count_by_cls=nms_by_class(boxes,0.4,method='linear',score_thres=score_thres) #  'gaussian'   ''   'original'   'linear'\n",
    "                boxes,count_by_cls=nms_by_class(boxes,0.4,method='',score_thres=score_thres)\n",
    "                num=len(boxes)\n",
    "                total_by_cls=dict(Counter(total_by_cls)+Counter(count_by_cls))\n",
    "                print('After nms_2 %s.json->total:%d'%(name,num))\n",
    "#     print('nms_all')\n",
    "#     boxes,count_by_cls=nms_all(boxes,0.4,method='linear',score_thres=score_thres) #  'gaussian'   ''   'original'   'linear'   \n",
    "#     num=len(boxes)\n",
    "#     total_by_cls=dict(Counter(total_by_cls)+Counter(count_by_cls))\n",
    "#     print('After nms_all %s.json->total:%d'%(name,num))\n",
    "    ############################################\n",
    "    total+=num\n",
    "    outjson=os.path.join(outpath,name+'.json')\n",
    "    with open(outjson,'w') as f:\n",
    "        json.dump(boxes,f)\n",
    "print('get nms outs, total: %d %s, cost time:%f'%(total,total_by_cls,time.time()-start))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#  返回原图坐标\n",
    "def get_out(results2):\n",
    "    start=time.time()\n",
    "    lost2=[]\n",
    "    out2={}\n",
    "    i=1\n",
    "    count=0\n",
    "    for res in results2:\n",
    "\n",
    "\n",
    "        d={}\n",
    "        id=res['image_id']\n",
    "        which=images_dict[id]\n",
    "        try:\n",
    "            d['x'] = which[0] + res['bbox'][0]\n",
    "            d['y'] = which[1] + res['bbox'][1]\n",
    "            d['w'] = res['bbox'][2]\n",
    "            d['h'] = res['bbox'][3]\n",
    "            d['p'] = res['score']\n",
    "            d['class'] = categories_dict[res['category_id']]\n",
    "            name = which[-1]\n",
    "            count+=1\n",
    "            if name in out2:\n",
    "                out2[name].append(d)\n",
    "    #             print(out)\n",
    "            else:\n",
    "                out2[name]=[d]\n",
    "    #             print('No.%d :Original %s'%(i,name))\n",
    "    #             print('e.g.',d)\n",
    "                i+=1\n",
    "        except:\n",
    "            name=name.split(\"_\")[1]\n",
    "            if name not in lost:\n",
    "                lost.append(name)\n",
    "                print(lost)\n",
    "            continue\n",
    "\n",
    "    print('get outs,total: %d, cost time:%f'%(count,time.time()-start))\n",
    "    \n",
    "    return  out2\n",
    "##1:'ASC-H', 2: 'ASC-US', 3: 'HSIL', 4: 'LSIL', 5: 'Candida', 6: 'Trichomonas'\n",
    "def suan(out_file,Threshold = -0.05):\n",
    "    n1 = 0\n",
    "    n2 = 0\n",
    "    n3 = 0\n",
    "    n4 = 0\n",
    "    n5 = 0\n",
    "    n6 = 0\n",
    "    for key in out_file:\n",
    "        for value in out_file[key]:\n",
    "            if value['p'] <= Threshold:    ###修改\n",
    "                continue\n",
    "            if value['class'] == 'ASC-H':\n",
    "                n1 += 1\n",
    "            elif value['class'] == 'ASC-US':\n",
    "                n2 += 1\n",
    "            elif value['class'] == 'HSIL':\n",
    "                n3 += 1\n",
    "            elif value['class'] == 'LSIL':\n",
    "                n4 += 1\n",
    "            elif value['class'] == 'Candida':\n",
    "                n5 += 1\n",
    "            elif value['class'] == 'Trichomonas':\n",
    "                n6 += 1\n",
    "            else:\n",
    "                print(\"cuo\")\n",
    "    print(n1,n2,n3,n4,n5,n6)\n",
    "def model_fusion_he (*args):\n",
    "    out_file = {}\n",
    "    out_name = []\n",
    "    for outA in args:\n",
    "        for name,bboxes in outA.items():\n",
    "            bbox1 = chuli5 (bboxes)\n",
    "            ###过滤0\n",
    "            bbox_s = []\n",
    "            for js1 in bbox1:\n",
    "                if js1['p']>0.05:\n",
    "                     bbox_s.append(js1) \n",
    "            if name not in out_name:\n",
    "                out_file[name] = bbox_s\n",
    "                out_name.append(name)\n",
    "            else:\n",
    "                for bbox in bbox_s:\n",
    "                    out_file[name].append(bbox)\n",
    "#     for name,bboxes in outB.items():\n",
    "#         bbox1 = chuli5 (bboxes)\n",
    "#         ###过滤0\n",
    "#         bbox_s = []\n",
    "#         for js1 in bbox1:\n",
    "#             if js1['p']>0.05:\n",
    "#                  bbox_s.append(js1) \n",
    "#         if name not in out_name:\n",
    "#             out_file[name] = bbox_s\n",
    "#             out_name.append(name)\n",
    "#         else:\n",
    "#             for bbox in bbox_s:\n",
    "#                 out_file[name].append(bbox)\n",
    "    \n",
    "    return out_file\n",
    "####融合\n",
    "def model_fusion_chuli (out_file):    \n",
    "    for name,bboxes in out_file.items():\n",
    "        tbboxes = bboxes\n",
    "        for i,box in enumerate( bboxes):  \n",
    "            pp = bboxes[i][\"p\"]\n",
    "#             if pp==0:      ###后面的p 为0的没有选举权，但是有投票权\n",
    "#                 continue\n",
    "            vote = []\n",
    "            for j,tbox in enumerate( tbboxes):\n",
    "                if i == j: # >=\n",
    "                    continue\n",
    "#                 if tbox['p'] == 0:\n",
    "#                     continue\n",
    "                if box['class'] != tbox['class']:\n",
    "                    continue\n",
    "                ##l两筐没关系\n",
    "                if box[\"x\"]+box['w']<=tbox[\"x\"] or box[\"y\"]+box['h']<=tbox[\"y\"] or box[\"x\"]>=tbox[\"x\"]+tbox[\"w\"] or box[\"y\"]>=tbox[\"y\"]+tbox[\"h\"]: \n",
    "                    continue\n",
    "                ####得到 iou   \n",
    "                in_h = min(box['y']+box['h'],tbox['y']+tbox['h'])-max(box['y'],tbox['y'])\n",
    "                in_w = min(box['x']+box['w'],tbox['x']+tbox['w'])-max(box['x'],tbox['x'])\n",
    "                union = box['w']*box['h']+tbox['w']*tbox['h']-in_h*in_w\n",
    "                iou = in_h*in_w/union\n",
    "                \n",
    "                if iou >0.4:  # 证明两者有关系\n",
    "                    vote.append(j)   \n",
    "#                 elif iou >0.3 and pp >=0.4:  # 证明两者有关系\n",
    "#                     vote.append(j)\n",
    "                else:\n",
    "                    pass\n",
    "            if len(vote)>=3:   #投票 的 决定  1时是 2票   \n",
    "#                 print(\"???\")\n",
    "                pass\n",
    "#                 for kk in vote:\n",
    "#                     bboxes[kk][\"p\"] = 0\n",
    "            else:\n",
    "                bboxes[i][\"p\"] = 0\n",
    "                    \n",
    "        \n",
    "        ##过滤 0 \n",
    "        bbox_s = []\n",
    "        for js1 in bboxes:\n",
    "            if js1['p']>0.05:\n",
    "                 bbox_s.append(js1)\n",
    "        out_file[name] = bbox_s\n",
    "    return out_file\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "out1 = get_out(results1)\n",
    "out2 = get_out(results2)\n",
    "out3 = get_out(results3)\n",
    "out4 = get_out(results4)\n",
    "out5 = get_out(results5)\n",
    "out6 = get_out(results6)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# tout = copy.deepcopy(out2)\n",
    "out_file = model_fusion_he (out2,out3,out4,out5,out6,out1)\n",
    "suan(out_file)\n",
    "# out_file = model_fusion_he (out_file,out1)\n",
    "# suan(out_file)\n",
    "\n",
    "out_file = model_fusion_chuli (out_file)\n",
    "suan(out_file)\n",
    "out_file = model_fusion_he (out_file)\n",
    "suan(out_file)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "##1:'ASC-H', 2: 'ASC-US', 3: 'HSIL', 4: 'LSIL', 5: 'Candida', 6: 'Trichomonas'\n",
    "n1 = 0\n",
    "n2 = 0\n",
    "n3 = 0\n",
    "n4 = 0\n",
    "n5 = 0\n",
    "n6 = 0\n",
    "for key in out_file:\n",
    "    for value in out_file[key]:\n",
    "        if value['p'] <= 0.05:    ###修改\n",
    "            continue\n",
    "        if value['class'] == 'ASC-H':\n",
    "            n1 += 1\n",
    "        elif value['class'] == 'ASC-US':\n",
    "            n2 += 1\n",
    "        elif value['class'] == 'HSIL':\n",
    "            n3 += 1\n",
    "        elif value['class'] == 'LSIL':\n",
    "            n4 += 1\n",
    "        elif value['class'] == 'Candida':\n",
    "            n5 += 1\n",
    "        elif value['class'] == 'Trichomonas':\n",
    "            n6 += 1\n",
    "        else:\n",
    "            print(\"cuo\")\n",
    "print(n1,n2,n3,n4,n5,n6)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 生成提交用json 文件\n",
    "outpath='./out_jsons/'\n",
    "if not os.path.exists(outpath):\n",
    "    os.makedirs(outpath)\n",
    "start=time.time()\n",
    "for name,bboxes in out_file.items():\n",
    "#     chuli (bboxes)\n",
    "    outjson=os.path.join(outpath,name+'.json')\n",
    "    with open(outjson,'w') as f:\n",
    "        json.dump(bboxes,f)\n",
    "print('get nms outs,p_non_zero , cost time:%f'%(time.time()-start))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#  检查 补全文件\n",
    "outpath='./out_jsons/'\n",
    "names = os.listdir(outpath)\n",
    "tests=glob.glob(outpath+'*.json')\n",
    "if os.path.exists(outpath+'.ipynb_checkpoints'):\n",
    "#     os.removedirs(outpath+'.ipynb_checkpoints')\n",
    "    names.remove('.ipynb_checkpoints')\n",
    "path1 = r\"/home/admin/jupyter/Data/test/\"\n",
    "tests=glob.glob(path1+'*.json')\n",
    "print(\"结果文件个数\",len(names),'should be ',len(tests))\n",
    "c=1\n",
    "for each in tests:              \n",
    "    name=each.split('/')[-1]\n",
    "    if name in names:\n",
    "        pass\n",
    "#         print('\\texist no.%d :%s'%(c,name))\n",
    "    else:\n",
    "        with open(outpath + name, \"w\") as f:\n",
    "            f.write('[]')\n",
    "        print('add no.%d :%s'%(c,name))\n",
    "    c+=1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "### 阴阳 分离\n",
    "import os\n",
    "import json\n",
    "path = \"./out_jsons_6/\"\n",
    "num = 0\n",
    "for name in os.listdir(path)[1:]:\n",
    "    with open(path+name,\"r\") as f:\n",
    "        js = json.load(f) \n",
    "    no = 0\n",
    "    yes1 = 0\n",
    "    yes2 = 0\n",
    "    for js1 in js:\n",
    "        if js1[\"class\"] ==\"Trichomonas\":\n",
    "            yes1 +=1\n",
    "        elif js1[\"class\"] ==\"Candida\":\n",
    "            yes2 +=1\n",
    "        else:\n",
    "            no +=1\n",
    "    print(name,\"异常\",no,\"滴虫\",yes1,\"念珠菌\",yes2)\n",
    "#     if no> yes1 and no > yes2:\n",
    "#         for i,js1 in enumerate(js):\n",
    "#             if js1[\"class\"] ==\"Trichomonas\":\n",
    "#                 js[i][\"p\"] =0\n",
    "#             elif js1[\"class\"] ==\"Candida\":\n",
    "#                 js[i][\"p\"] = 0\n",
    "#             else:\n",
    "#                 pass\n",
    "#         num += no\n",
    "                \n",
    "#     if yes1>no*4 and yes1 > yes2:\n",
    "#         for i,js1 in enumerate(js):\n",
    "#             if js1[\"class\"] ==\"Trichomonas\":\n",
    "#                 pass\n",
    "#             elif js1[\"class\"] ==\"Candida\":\n",
    "#                 js[i][\"p\"] = 0\n",
    "#             else:\n",
    "#                 js[i][\"p\"] = 0\n",
    "#         num += yes1\n",
    "#     if yes2>no and yes2 > yes1:\n",
    "#         for i,js1 in enumerate(js):\n",
    "#             if js1[\"class\"] ==\"Trichomonas\":\n",
    "#                 js[i][\"p\"] = 0  \n",
    "#             elif js1[\"class\"] ==\"Candida\":\n",
    "#                 pass\n",
    "#             else:\n",
    "#                 js[i][\"p\"] = 0\n",
    "#         num += yes2\n",
    "        \n",
    "#     new_js = []\n",
    "#     for js1 in js:\n",
    "#         if js1['p']>0.05:   ####自动 过滤\n",
    "#              new_js.append(js1) \n",
    "#     with open(path+name,\"w\") as f:\n",
    "#         json.dump(new_js,f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "##1:'ASC-H', 2: 'ASC-US', 3: 'HSIL', 4: 'LSIL', 5: 'Candida', 6: 'Trichomonas'\n",
    "n1 = 0\n",
    "n2 = 0\n",
    "n3 = 0\n",
    "n4 = 0\n",
    "n5 = 0\n",
    "n6 = 0\n",
    "path = \"./out_jsons/\"\n",
    "for name in os.listdir(path)[1:]:\n",
    "    with open(path+name,\"r\") as f:\n",
    "        js = json.load(f) \n",
    "    if len(js)==0:\n",
    "        continue\n",
    "    for value in js:\n",
    "        if value['class'] == 'ASC-H':\n",
    "            n1 += 1\n",
    "        elif value['class'] == 'ASC-US':\n",
    "            n2 += 1\n",
    "        elif value['class'] == 'HSIL':\n",
    "            n3 += 1\n",
    "        elif value['class'] == 'LSIL':\n",
    "            n4 += 1\n",
    "        elif value['class'] == 'Candida':\n",
    "            n5 += 1\n",
    "        elif value['class'] == 'Trichomonas':\n",
    "            n6 += 1\n",
    "        else:\n",
    "            print(\"cuo\")\n",
    "print(n1,n2,n3,n4,n5,n6)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}