bakeup

contrast/feat_analysisi.py

@@ -0,0 +1,160 @@
+# -*- coding: utf-8 -*-
+"""
+Created on Wed Nov 20 11:17:29 2024
+
+@author: ym
+"""
+
+import os
+import cv2
+import pickle
+import numpy as np
+from scipy.spatial.distance import cdist
+import matplotlib.pyplot as plt
+
+def save_imgpairs(barcode, imgpaths, matrix, savepath, thresh=(0.4, 0.6), ctype="intra"):
+    if ctype=="intra":
+        rows, cols = np.triu_indices(matrix.shape[0], k=1)   # k=1 表示不包括对角线
+        mask = matrix[rows, cols] < thresh[1]
+        indices = list(zip(rows[mask], cols[mask]))
+    else:
+        rows, cols = np.where(matrix > thresh[0])
+        indices = list(zip(rows, cols))
+    
+    
+    if len(indices):
+        savepath = os.path.join(savepath, barcode)
+        if not os.path.exists(savepath):
+            os.makedirs (savepath)
+    
+    
+    for idx1, idx2 in indices:
+        if len(imgpaths) == 1:
+            img1 = cv2.imread(imgpaths[0][idx1])
+            img2 = cv2.imread(imgpaths[0][idx2])
+        elif len(imgpaths) == 2:
+            img1 = cv2.imread(imgpaths[0][idx1])
+            img2 = cv2.imread(imgpaths[1][idx2])
+            
+            
+        
+        simi = matrix[idx1, idx2]
+
+        H1, W1 = img1.shape[:2]
+        H2, W2 = img2.shape[:2]
+        H, W = max((H1, H2)), max((W1, W2))
+        img = np.ones((H, 2*W, 3), dtype=np.uint8) *np.array([255, 128, 128])
+        
+        img[0:H1, 0:W1, :] = img1
+        img[0:H2, (2*W-W2):, :] = img2
+        
+        text = f"sim: {simi:.2f}"
+        org = (10, H-10)
+        cv2.putText(img, text, org, fontFace=cv2.FONT_HERSHEY_SIMPLEX, fontScale=0.75, 
+                    color=(0, 0, 255), thickness=2, lineType=cv2.LINE_AA)
+        imgpath = os.path.join(savepath, f"{simi:.2f}_{barcode}_{idx1}_{idx2}.png")
+        cv2.imwrite(imgpath, img)
+        
+
+
+def feat_analysis(featpath):    
+    savepath = r"D:\exhibition\result\stdfeat"
+
+    InterThresh = (0.4, 0.6)
+    
+    featDict, features= [], []
+    for filename in os.listdir(featpath):
+        file, ext = os.path.splitext(filename)
+        if ext != ".pickle": continue
+        
+        filepath = os.path.join(featpath, filename)
+        with open(filepath, 'rb') as f:  
+            bpDict = pickle.load(f)
+
+            feat = bpDict["feats_ft32"]
+
+            featDict.append(bpDict)
+            features.append(feat)
+    N = len(features)   
+   
+    simMatrix = []
+    intra_simi = np.empty(0)   
+    low_simi_index = {}
+    for i, feats in enumerate(features):
+        matrix = 1 - cdist(feats, feats, 'cosine')
+        
+        simMatrix.append(matrix)
+        
+        '''提取相似矩阵上三角元素'''
+        rows, cols = np.triu_indices(matrix.shape[0], k=1)  # k=1 表示不包括对角线
+        upper_tri= matrix[rows, cols]        
+        intra_simi = np.concatenate((intra_simi, upper_tri))
+        
+        '''保存相似度小于阈值的图像对'''
+        barcode = featDict[i]["barcode"]
+        imgpaths = featDict[i]["imgpaths"]
+        # save_imgpairs(barcode, [imgpaths], matrix, savepath, InterThresh, "intra")
+        print(f"{barcode} have done!")
+        
+    Matrix = np.zeros((N, N))
+    inter_bcds = []
+    inter_simi = np.empty(0)
+    for i, feati in enumerate(features):
+        bcdi = featDict[i]["barcode"]
+        imgpathi = featDict[i]["imgpaths"]
+        for j, featj in enumerate(features):
+            bcdj = featDict[j]["barcode"]
+            imgpathj = featDict[j]["imgpaths"]
+        
+            matrix = 1 - cdist(feati, featj, 'cosine')
+            
+            inter_bcds.append((i, j, bcdi, bcdj))
+            Matrix[i, j] = np.mean(matrix)
+            if j>i:
+                bcd_ij = bcdi+'_'+bcdj
+                # save_imgpairs(bcd_ij, [imgpathi, imgpathj], matrix, savepath, InterThresh, "inter")
+                inter_simi = np.concatenate((inter_simi, matrix.ravel()))
+                
+                print(f"{bcd_ij} have done!")
+            
+    fig, axs = plt.subplots(2, 1)
+    axs[0].hist(intra_simi, bins=100, color='blue', edgecolor='black', alpha=0.7)
+    axs[0].set_xlim(0, 1)
+    axs[0].set_xlabel('Performance')
+    axs[0].set_title("inter similarity")
+    
+    axs[1].hist(inter_simi, bins=100, color='green', edgecolor='black', alpha=0.7)
+    axs[1].set_xlim(0, 1)
+    axs[1].set_xlabel('Performance')
+    axs[1].set_title("inter similarity")
+    
+    
+    
+    
+    
+    print("Done！")
+            
+            
+            
+    
+
+
+
+
+def main():
+    stdpath = r"D:\exhibition\dataset\feats"
+    
+    feat_analysis(stdpath)
+    
+
+if __name__ == '__main__':
+    main()
+    
+    
+    
+    
+    
+    
+    
+    
+    

contrast/feat_extract/config.py

@@ -10,6 +10,8 @@ class Config:
     embedding_size = 256
     drop_ratio = 0.5
     img_size = 224
+    
+    batch_size = 8
 
     # data preprocess
     # input_shape = [1, 128, 128]
@@ -58,7 +60,11 @@ class Config:
     testbackbone = 'resnet18'  # [resnet18, mobilevit_s, mobilenet_v2, mobilenetv3_small, mobilenetv3_large, mobilenet_v1, PPLCNET_x1_0, PPLCNET_x0_5]
     test_val = "D:/比对/cl"
     # test_val = "./data/test_data_100"
-    test_model = "checkpoints/resnet18_0515/best.pth"
+    
+    # test_model = "checkpoints/best_resnet18_v11.pth"   
+    test_model = "checkpoints/zhanting_cls22_v11.pth"
+    
+    
 
     train_batch_size = 512  # 256
     test_batch_size = 256  # 256

contrast/feat_extract/inference.py

@@ -3,30 +3,140 @@
 
 @author: LiChen
 """
-
-
+# import pdb
+# import shutil
+import torch.nn as nn
+# import statistics
 import os
-import os.path as osp
-import pdb
 import numpy as np 
-import shutil
 from scipy.spatial.distance import cdist
 import torch
-import torch.nn as nn
+import os.path as osp
 from PIL import Image
 import json
-from config import config as conf
-from model import resnet18
+import matplotlib.pyplot as plt
+from pathlib import Path
+# import sys
+# sys.path.append(r"D:\DetectTracking")
+# from contrast.config import config as conf
+# from contrast.model import resnet18
+
+from .config import config as conf
+from .model import resnet18
 
 # from model import (mobilevit_s, resnet14, resnet18, resnet34, resnet50, mobilenet_v2,
 #                    MobileNetV3_Small, mobilenet_v1, PPLCNET_x1_0, PPLCNET_x0_5, PPLCNET_x2_5)
 
+curpath = Path(__file__).resolve().parents[0]
+
+class FeatsInterface:
+    def __init__(self, conf):
+        self.device = conf.device
+
+        # if conf.backbone == 'resnet18':
+        #     model = resnet18().to(conf.device)
+            
+        model = resnet18().to(conf.device)
+        self.transform = conf.test_transform
+        self.batch_size = conf.batch_size
+        self.embedding_size = conf.embedding_size
+        
+        if conf.test_model.find("zhanting") == -1:
+            model = nn.DataParallel(model).to(conf.device)
+        self.model = model
+        
+        modpath = os.path.join(curpath, conf.test_model)
+        self.model.load_state_dict(torch.load(modpath, map_location=conf.device))
+        self.model.eval()
+        print('load model {} '.format(conf.testbackbone))
+    
+    def inference(self, images, detections=None):
+        '''
+        如果是BGR，需要转变为RGB格式
+        '''
+        if isinstance(images, np.ndarray):
+            imgs, features = self.inference_image(images, detections)
+            return imgs, features
+                  
+        batch_patches = []
+        patches = []
+        for i, img in enumerate(images):
+            img = img.copy()   
+            patch = self.transform(img)            
+            if str(self.device) != "cpu":
+                patch = patch.to(device=self.device).half()
+            else:
+                patch = patch.to(device=self.device)
+            
+            patches.append(patch)
+            if (i + 1) % self.batch_size == 0:
+                patches = torch.stack(patches, dim=0)
+                batch_patches.append(patches)
+                patches = []
+        
+        if len(patches):
+            patches = torch.stack(patches, dim=0)
+            batch_patches.append(patches)
+        
+        features = np.zeros((0, self.embedding_size))
+        for patches in batch_patches:
+            pred=self.model(patches)
+            pred[torch.isinf(pred)] = 1.0            
+            feat = pred.cpu().data.numpy()
+            features = np.vstack((features, feat))        
+        return features
+    
+    def inference_image(self, image, detections):
+        H, W, _ = np.shape(image)
+
+        batch_patches = []
+        patches = []
+        imgs = []
+        for d in range(np.size(detections, 0)):
+            tlbr = detections[d, :4].astype(np.int_)
+            tlbr[0] = max(0, tlbr[0])
+            tlbr[1] = max(0, tlbr[1])
+            tlbr[2] = min(W - 1, tlbr[2])
+            tlbr[3] = min(H - 1, tlbr[3])
+            img = image[tlbr[1]:tlbr[3], tlbr[0]:tlbr[2], :]
+            
+            imgs.append(img)
+            
+            
+            img1 = img[:, :, ::-1].copy()    # the model expects RGB inputs
+            patch = self.transform(img1)
+
+            # patch = patch.to(device=self.device).half()
+            if str(self.device) != "cpu":
+                patch = patch.to(device=self.device).half()
+            else:
+                patch = patch.to(device=self.device)
+            
+            patches.append(patch)
+            if (d + 1) % self.batch_size == 0:
+                patches = torch.stack(patches, dim=0)
+                batch_patches.append(patches)
+                patches = []
+
+        if len(patches):
+            patches = torch.stack(patches, dim=0)
+            batch_patches.append(patches)
+
+        features = np.zeros((0, self.embedding_size))
+        for patches in batch_patches:
+            pred = self.model(patches)
+            pred[torch.isinf(pred)] = 1.0
+            feat = pred.cpu().data.numpy()
+            features = np.vstack((features, feat))
+
+        return imgs, features 
+        
+        
+    
+    
+
+
 
-import matplotlib.pyplot as plt
-import statistics
-embedding_size = conf.embedding_size
-img_size = conf.img_size
-device = conf.device
 
 
 def unique_image(pair_list) -> set:
@@ -102,38 +212,38 @@ def featurize(images: list, transform, net, device, train=False) -> dict:
         res = {img: feature for (img, feature) in zip(images, features)}
     return res
 
-def inference_image(images: list, transform, net, device, bs=16, embedding_size=256) -> dict:
-    batch_patches = []
-    patches = []
-    for d, img in enumerate(images):
-        img = Image.open(img)
-        patch = transform(img)
+# def inference_image(images: list, transform, net, device, bs=16, embedding_size=256) -> dict:
+#     batch_patches = []
+#     patches = []
+#     for d, img in enumerate(images):
+#         img = Image.open(img)
+#         patch = transform(img)
         
-        if str(device) != "cpu":
-            patch = patch.to(device).half()
-        else:
-            patch = patch.to(device)
+#         if str(device) != "cpu":
+#             patch = patch.to(device).half()
+#         else:
+#             patch = patch.to(device)
         
-        patches.append(patch)
-        if (d + 1) % bs == 0:
-            patches = torch.stack(patches, dim=0)
-            batch_patches.append(patches)
-            patches = []
+#         patches.append(patch)
+#         if (d + 1) % bs == 0:
+#             patches = torch.stack(patches, dim=0)
+#             batch_patches.append(patches)
+#             patches = []
 
-    if len(patches):
-        patches = torch.stack(patches, dim=0)
-        batch_patches.append(patches)
+#     if len(patches):
+#         patches = torch.stack(patches, dim=0)
+#         batch_patches.append(patches)
 
-    features = np.zeros((0, embedding_size), dtype=np.float32)
-    for patches in batch_patches:
-        pred = net(patches)
-        pred[torch.isinf(pred)] = 1.0
-        feat = pred.cpu().data.numpy()
-        features = np.vstack((features, feat))
+#     features = np.zeros((0, embedding_size), dtype=np.float32)
+#     for patches in batch_patches:
+#         pred = net(patches)
+#         pred[torch.isinf(pred)] = 1.0
+#         feat = pred.cpu().data.numpy()
+#         features = np.vstack((features, feat))
         
     
 
-    return features
+#     return features
 
 
 
@@ -283,6 +393,7 @@ def compute_contrast_accuracy(content_list_read):
     
     npairs = min((len(same_folder_pairs), len(cross_folder_pairs)))
     
+    Encoder = FeatsInterface(conf)
     
     same_pairs = same_folder_pairs[:npairs]
     cross_pairs = cross_folder_pairs[:npairs]
@@ -292,8 +403,8 @@ def compute_contrast_accuracy(content_list_read):
         images_a = [osp.join(conf.test_val, img) for img in same_pairs[i][0]]
         images_b = [osp.join(conf.test_val, img) for img in same_pairs[i][1]]
         
-        feats_a = inference_image(images_a, conf.test_transform, model, conf.device)
-        feats_b = inference_image(images_b, conf.test_transform, model, conf.device)
+        feats_a = Encoder.inference(images_a)
+        feats_b = Encoder.inference(images_b)
         # matrix = 1- np.maximum(0.0, cdist(feats_a, feats_b, 'cosine'))
         matrix = 1 - cdist(feats_a, feats_b, 'cosine')
         
@@ -324,8 +435,8 @@ def compute_contrast_accuracy(content_list_read):
         images_a = [osp.join(conf.test_val, img) for img in cross_pairs[i][0]]
         images_b = [osp.join(conf.test_val, img) for img in cross_pairs[i][1]]
         
-        feats_a = inference_image(images_a, conf.test_transform, model, conf.device)
-        feats_b = inference_image(images_b, conf.test_transform, model, conf.device)
+        feats_a = Encoder.inference(images_a)
+        feats_b = Encoder.inference(images_b)
         # matrix = 1- np.maximum(0.0, cdist(feats_a, feats_b, 'cosine'))
         matrix = 1 - cdist(feats_a, feats_b, 'cosine')
         
@@ -407,28 +518,28 @@ if __name__ == '__main__':
 
     # Network Setup
     if conf.testbackbone == 'resnet18':
-        # model = ResIRSE(img_size, embedding_size, conf.drop_ratio).to(device)
-        model = resnet18().to(device)
+        # model = ResIRSE(conf.img_size, conf.embedding_size, conf.drop_ratio).to(conf.device)
+        model = resnet18().to(conf.device)
     # elif conf.testbackbone == 'resnet34':
-    #     model = resnet34().to(device)
+    #     model = resnet34().to(conf.device)
     # elif conf.testbackbone == 'resnet50':
-    #     model = resnet50().to(device)
+    #     model = resnet50().to(conf.device)
     # elif conf.testbackbone == 'mobilevit_s':
-    #     model = mobilevit_s().to(device)
+    #     model = mobilevit_s().to(conf.device)
     # elif conf.testbackbone == 'mobilenetv3':
-    #     model = MobileNetV3_Small().to(device)
+    #     model = MobileNetV3_Small().to(conf.device)
     # elif conf.testbackbone == 'mobilenet_v1':
-    #     model = mobilenet_v1().to(device)
+    #     model = mobilenet_v1().to(conf.device)
     # elif conf.testbackbone == 'PPLCNET_x1_0':
-    #     model = PPLCNET_x1_0().to(device)
+    #     model = PPLCNET_x1_0().to(conf.device)
     # elif conf.testbackbone == 'PPLCNET_x0_5':
-    #     model = PPLCNET_x0_5().to(device)
+    #     model = PPLCNET_x0_5().to(conf.device)
     # elif conf.backbone == 'PPLCNET_x2_5':
-    #     model = PPLCNET_x2_5().to(device)
+    #     model = PPLCNET_x2_5().to(conf.device)
     # elif conf.testbackbone == 'mobilenet_v2':
-    #     model = mobilenet_v2().to(device)
+    #     model = mobilenet_v2().to(conf.device)
     # elif conf.testbackbone == 'resnet14':
-    #     model = resnet14().to(device)
+    #     model = resnet14().to(conf.device)
     else:
         raise ValueError('Have not model {}'.format(conf.backbone))
 

contrast/feat_extract/model/mobilenet_v2.py

@@ -1,6 +1,6 @@
 from torch import nn
 from .utils import load_state_dict_from_url
-from config import config as conf
+from ..config import config as conf
 
 __all__ = ['MobileNetV2', 'mobilenet_v2']
 

contrast/feat_extract/model/mobilenet_v3.py

@@ -7,7 +7,7 @@ import torch
 import torch.nn as nn
 import torch.nn.functional as F
 from torch.nn import init
-from config import config as conf
+from ..config import config as conf
 
 
 class hswish(nn.Module):

contrast/feat_extract/model/mobilevit.py

@@ -2,7 +2,10 @@ import torch
 import torch.nn as nn
 
 from einops import rearrange
-from config import config as conf
+
+# import sys
+# sys.path.append(r"D:\DetectTracking")
+from ..config import config as conf
 
 
 def conv_1x1_bn(inp, oup):

contrast/feat_extract/model/resbam.py

@@ -1,7 +1,10 @@
-from model.CBAM import CBAM
 import torch
 import torch.nn as nn
-from model.Tool import GeM as gem
+
+from .CBAM import CBAM
+from .Tool import GeM as gem
+# from model.CBAM import CBAM
+# from model.Tool import GeM as gem
 
 
 class Bottleneck(nn.Module):

contrast/feat_extract/model/resnet_pre.py

@@ -1,6 +1,6 @@
 import torch
 import torch.nn as nn
-from config import config as conf
+from ..config import config as conf
 
 try:
     from torch.hub import load_state_dict_from_url

contrast/genfeats.py

@@ -6,26 +6,33 @@ Created on Sun Nov  3 12:05:19 2024
 """
 import os
 import time
-import torch
+# import torch
 import pickle
+# import json
 import numpy as np
-from config import config as conf
-from model import resnet18 as resnet18
-from feat_inference import inference_image
+from PIL import Image
+from feat_extract.config import config as conf
+# from model import resnet18 as resnet18
+from feat_extract.inference import FeatsInterface #, inference_image
 
 
 IMG_FORMAT = ['.bmp', '.jpg', '.jpeg', '.png']
 
-'''======= 0. 配置特征提取模型地址 ======='''
-model_path = conf.test_model
-model_path = r"D:\exhibition\ckpt\zhanting.pth"
+# def model_init(conf, mpath=None):
+#     '''======= 0. 配置特征提取模型地址 ======='''
+#     if mpath is None:
+#         model_path = conf.test_model
+#     else:
+#         model_path = mpath
 
-##============ load resnet mdoel 
-model = resnet18().to(conf.device)
-# model = nn.DataParallel(model).to(conf.device)
-model.load_state_dict(torch.load(model_path, map_location=conf.device))
-model.eval()
-print('load model {} '.format(conf.testbackbone))
+#     ##============ load resnet mdoel 
+#     model = resnet18().to(conf.device)
+#     # model = nn.DataParallel(model).to(conf.device)
+#     model.load_state_dict(torch.load(model_path, map_location=conf.device))
+#     model.eval()
+#     print('load model {} '.format(conf.testbackbone))
+
+#     return model
 
 def get_std_barcodeDict(bcdpath, savepath, bcdSet):
     '''
@@ -42,9 +49,9 @@ def get_std_barcodeDict(bcdpath, savepath, bcdSet):
     '''读取数据集中 barcode 列表'''
     stdBarcodeList = []
     for filename in os.listdir(bcdpath):
-        # filepath = os.path.join(bcdpath, filename)
-        # if not os.path.isdir(filepath) or not filename.isdigit() or len(filename)<8: 
-        #     continue
+        filepath = os.path.join(bcdpath, filename)
+        if not os.path.isdir(filepath) or not filename.isdigit() or len(filename)<8: 
+            continue
         if bcdSet is None:
             stdBarcodeList.append(filename)
         elif filename in bcdSet:
@@ -59,7 +66,7 @@ def get_std_barcodeDict(bcdpath, savepath, bcdSet):
     for barcode, bpath in bcdPaths:  
         pickpath = os.path.join(savepath, f"{barcode}.pickle")
         if os.path.isfile(pickpath):
-            continue
+            continue 
         
         stdBarcodeDict = {}
         stdBarcodeDict[barcode] = []
@@ -89,6 +96,7 @@ def get_std_barcodeDict(bcdpath, savepath, bcdSet):
         pickpath = os.path.join(savepath, f"{barcode}.pickle")
         with open(pickpath, 'wb') as f:
             pickle.dump(stdBarcodeDict, f)
+
         print(f"Barcode: {barcode}")
 
         # k += 1
@@ -115,32 +123,37 @@ def stdfeat_infer(imgPath, featPath, bcdSet=None):
     stdBarcodeDict = {}
     stdBarcodeDict_ft16 = {}
     
+    Encoder = FeatsInterface(conf)
     
-    '''4处同名: (1)barcode原始图像文件夹; (2)imgPath中的 .pickle 文件名、该pickle文件中字典的key值'''
-    
+    '''4处同名: (1)barcode原始图像文件夹; (2)imgPath中的 .pickle 文件名; 
+               (3)该pickle文件中字典的key值; (4)特征向量字典中的一个key值'''
     k = 0
     for filename in os.listdir(imgPath):
         bcd, ext = os.path.splitext(filename)
-        pkpath = os.path.join(featPath, f"{bcd}.pickle")
-        
-        if os.path.isfile(pkpath): continue
+        filepath = os.path.join(imgPath, filename)
+        if ext != ".pickle": continue
         if bcdSet is not None and bcd not in bcdSet:
             continue
 
-        filepath = os.path.join(imgPath, filename)
+        featpath = os.path.join(featPath, f"{bcd}.pickle")
 
         stdbDict = {}
-        stdbDict_ft16 = {}
-        stdbDict_uint8 = {}
-        
         t1 = time.time()
-        
         try:
             with open(filepath, 'rb') as f:  
-                bpDict = pickle.load(f)    
+                bpDict = pickle.load(f)   
+ 
             for barcode, imgpaths in bpDict.items():
                 # feature = batch_inference(imgpaths, 8) #from vit distilled model of LiChen
-                feature = inference_image(imgpaths, conf.test_transform, model, conf.device)
+                # feature = inference_image(imgpaths, conf.test_transform, model, conf.device)
+                
+                imgs = []
+                for d, imgpath in enumerate(imgpaths):
+                    img = Image.open(imgpath)
+                    imgs.append(img)
+                
+                feature = Encoder.inference(imgs)
+                
                 feature /= np.linalg.norm(feature, axis=1)[:, None]
                 
                 # float16
@@ -162,7 +175,7 @@ def stdfeat_infer(imgPath, featPath, bcdSet=None):
         stdbDict["feats_ft16"] = feature_ft16
         stdbDict["feats_uint8"] = feature_uint8
         
-        with open(pkpath, 'wb') as f:
+        with open(featpath, 'wb') as f:
             pickle.dump(stdbDict, f)
         
         stdBarcodeDict[barcode] = feature
@@ -174,21 +187,10 @@ def stdfeat_infer(imgPath, featPath, bcdSet=None):
         # if k == 10:
         #     break
     
-    ##================== float32 
-    # pickpath = os.path.join(featPath, f"barcode_features_{k}.pickle")
-    # with open(pickpath, 'wb') as f:
-    #     pickle.dump(stdBarcodeDict, f)  
-       
-    ##================== float16   
-    # pickpath_ft16 = os.path.join(featPath, f"barcode_features_ft16_{k}.pickle")
-    # with open(pickpath_ft16, 'wb') as f:
-    #     pickle.dump(stdBarcodeDict_ft16, f)  
-    
     return
 
 
-
-def genfeatures(imgpath, bcdpath, featpath, bcdSet=None):
+def gen_bcd_features(imgpath, bcdpath, featpath, bcdSet=None):
     ''' 生成标准特征集 '''
     '''1. 提取 imgpath 中样本地址，生成字典{barcode: [imgpath1, imgpath1, ...]}
           并存储于: bcdpath, 格式为 barcode.pickle'''
@@ -198,11 +200,12 @@ def genfeatures(imgpath, bcdpath, featpath, bcdSet=None):
     stdfeat_infer(bcdpath, featpath, bcdSet)
     
 def main():
-    imgpath = r"\\192.168.1.28\share\展厅barcode数据\整理\zhantingBase"
+    imgpath = r"\\192.168.1.28\share\数据\已完成数据\展厅数据\v1.0\比对数据\整理\zhantingBase"
     bcdpath = r"D:\exhibition\dataset\bcdpath"  
     featpath = r"D:\exhibition\dataset\feats"
-    
-    genfeatures(imgpath, bcdpath, featpath)
+
+
+    gen_bcd_features(imgpath, bcdpath, featpath)
     
 
     

contrast/one2n_contrast.py

@@ -325,7 +325,7 @@ def one2n_deleted(all_list):
 
 
 
-def one2n_return(all_list, basepath): 
+def one2n_return(all_list): 
     corrpairs, corr_similarity, errpairs, err_similarity = [], [], [], []
 
     for s_list in all_list:
@@ -410,7 +410,7 @@ def test_rpath_return():
     savepath = r'D:\DetectTracking\contrast\result' 
     
     all_list = read_returnGoods_file(return_bfile)
-    corrpairs, errpairs, _, _ = one2n_return(all_list, basepath)
+    corrpairs, errpairs, _, _ = one2n_return(all_list)
     for corrpair in corrpairs:
         GetoutPath = os.path.join(basepath, corrpair[0])
         InputPath = os.path.join(basepath, corrpair[1])
@@ -435,7 +435,7 @@ def test_one2n():
     savepath: pr曲线保存路径 
     '''
     # fpath = r'\\192.168.1.28\share\测试_202406\deletedBarcode\other'        # deletedBarcode.txt
-    fpath = r'\\192.168.1.28\share\测试_202406\returnGoods\all'               # returnGoods.txt
+    fpath = r'\\192.168.1.28\share\测试_202406\1108_展厅模型v800测试'               # returnGoods.txt
     savepath = r'\\192.168.1.28\share\测试_202406\deletedBarcode\illustration'
 
     if os.path.isdir(fpath):
@@ -476,9 +476,9 @@ def test_one2n():
         plt1.savefig(os.path.join(savepath, file+'_pr.png'))
         # plt1.close()
         
-        # plt2 = showHist(err_similarity, correct_similarity)
-        # plt2.show()
-        # plt2.savefig(os.path.join(savepath, file+'_hist.png'))
+        plt2 = showHist(err_similarity, correct_similarity)
+        plt2.show()
+        plt2.savefig(os.path.join(savepath, file+'_hist.png'))
         # plt.close()
 
 
@@ -486,7 +486,7 @@ def test_one2n():
 if __name__ == '__main__':
     # test_one2n()
     test_rpath_return()     # returnGoods.txt
-    test_rpath_deleted()    # deleteBarcode.txt
+    # test_rpath_deleted()    # deleteBarcode.txt
     
     
     # try:

contrast/one2one_contrast.py

@@ -49,7 +49,8 @@ from datetime import datetime
 sys.path.append(r"D:\DetectTracking")
 from tracking.utils.read_data import extract_data, read_tracking_output, read_one2one_simi, read_deletedBarcode_file
 
-from genfeats import genfeatures, stdfeat_infer
+from config import config as conf
+from genfeats import model_init, genfeatures, stdfeat_infer
 
 IMG_FORMAT = ['.bmp', '.jpg', '.jpeg', '.png']
 
@@ -546,10 +547,10 @@ def test_one2one():
         
     
     
-    
+    model = model_init(conf)
     
     '''==== 1. 生成标准特征集, 只需运行一次 ==============='''    
-    genfeatures(stdSamplePath, stdBarcodePath, stdFeaturePath, bcdSet)
+    genfeatures(model, stdSamplePath, stdBarcodePath, stdFeaturePath, bcdSet)
     print("stdFeats have generated and saved!")
     
     

contrast/one2one_onsite.py

@@ -7,7 +7,12 @@ Created on Wed Sep 11 11:57:30 2024
 """
 import os
 import numpy as np
+from pathlib import Path
 import matplotlib.pyplot as plt
+import sys
+
+sys.path.append(r"D:\DetectTracking")
+from tracking.utils.read_data import read_similar
 
 def read_one2one_data(filepath):
     simiList = []
@@ -85,7 +90,7 @@ def plot_pr_curve(matrix):
     pass
 
                 
-def main():
+def test_compare():
     filepaths = [r"\\192.168.1.28\share\测试_202406\0913_扫A放B\0913_1\OneToOneCompare.txt",
                  r"\\192.168.1.28\share\测试_202406\0913_扫A放B\0913_2\OneToOneCompare.txt",
                  r"\\192.168.1.28\share\测试_202406\0914_扫A放B\0914_1\OneToOneCompare.txt",
@@ -99,10 +104,251 @@ def main():
 
     plot_pr_curve(simiList)
     
+def one2one_pr(paths): 
+    paths = Path(paths)
+    evtpaths = [p for p in paths.iterdir() if p.is_dir() and len(p.name.split('_'))>=2]
+    
+    events, similars = [], []
+    
+    ##===================================== 扫A放A, 扫A放B场景
+    one2oneAA, one2oneAB = [], []
+    
+    ##===================================== 应用于展厅 1：N
+    tp_events, fn_events, fp_events, tn_events = [], [], [], []
+    tp_simi, fn_simi, tn_simi, fp_simi = [], [], [], []
+    
+    ##===================================== 应用于1:n
+    tpevents, fnevents, fpevents, tnevents = [], [], [], []
+    tpsimi, fnsimi, tnsimi, fpsimi = [], [], [], []
+
+    for path in evtpaths:
+        barcode = path.stem.split('_')[-1]
+        datapath = path.joinpath('process.data')
+        
+        if not barcode.isdigit() or len(barcode)<10: continue
+        if not datapath.is_file(): continue
+    
+        try: 
+            SimiDict = read_similar(datapath)
+        except Exception as e:
+            print(f"{path.stem}, Error: {e}")
+        
+        one2one = SimiDict['one2one']
+        one2n = SimiDict['one2n']
+        
+        barcodes, similars = [], []
+        for dt in one2one:
+            barcodes.append(dt['barcode'])
+            similars.append(dt['similar'])
+        
+        if len(barcodes)!=len(similars) or len(barcodes)==0:
+            continue
+            
+        ##===================================== 扫A放A, 扫A放B场景    
+        simAA = [similars[i] for i in range(len(barcodes)) if barcodes[i]==barcode] 
+        simAB = [similars[i] for i in range(len(barcodes)) if barcodes[i]!=barcode]
+        
+        one2oneAA.extend(simAA)
+        one2oneAB.extend(simAB)
+        
+        
+        ##===================================== 以下应用适用于展厅 1：N
+        max_idx = similars.index(max(similars))
+        max_sim = similars[max_idx]
+        # max_bcd = barcodes[max_idx]
+        
+        for i in range(len(one2one)):
+            bcd, simi = barcodes[i], similars[i]
+            if bcd==barcode and simi==max_sim:
+                tp_simi.append(simi)
+                tp_events.append(path.stem)
+            elif bcd==barcode and simi!=max_sim:
+                fn_simi.append(simi)
+                fn_events.append(path.stem)
+            elif bcd!=barcode and simi!=max_sim: 
+                tn_simi.append(simi)
+                tn_events.append(path.stem)
+            else:
+                fp_simi.append(simi)
+                fp_events.append(path.stem)
+        
+                
+        ##===================================== 以下应用适用1:n 
+        events, evt_barcodes, evt_similars, evt_types = [], [], [], []        
+        for dt in one2n:
+            events.append(dt["event"])
+            evt_barcodes.append(dt["barcode"])
+            evt_similars.append(dt["similar"])
+            evt_types.append(dt["type"])
+        
+        if len(events)!=len(evt_barcodes) or len(evt_barcodes)!=len(evt_similars) \
+            or len(evt_barcodes)!=len(evt_similars) or len(events)==0: continue 
+  
+        maxsim = evt_similars[evt_similars.index(max(evt_similars))]
+        for i in range(len(one2n)):
+            bcd, simi = evt_barcodes[i], evt_similars[i]
+            
+            if bcd==barcode and simi==maxsim:
+                tpsimi.append(simi)
+                tpevents.append(path.stem)
+            elif bcd==barcode and simi!=maxsim:
+                fnsimi.append(simi)
+                fnevents.append(path.stem)
+            elif bcd!=barcode and simi!=maxsim: 
+                tnsimi.append(simi)
+                tnevents.append(path.stem)
+            else:
+                fpsimi.append(simi)
+                fpevents.append(path.stem)
+            
+    '''命名规则:
+          1:1              1:n                1:N
+          TP_              TP                 TPX
+        PPrecise_       PPrecise           PPreciseX
+                         tpsimi             tp_simi  
+    '''
+    
+    ''' 1:1 数据存储'''
+    PPrecise_, PRecall_ = [], []
+    NPrecise_, NRecall_ = [], []
+    
+    ''' 1:n 数据存储'''
+    PPrecise, PRecall = [], []
+    NPrecise, NRecall = [], []
+    
+    ''' 展厅 1:N 数据存储'''
+    PPreciseX, PRecallX = [], []
+    NPreciseX, NRecallX = [], []
+    
+    Thresh = np.linspace(-0.2, 1, 100)
+    for th in Thresh:
+        '''============================= 1:1'''
+        TP_ = sum(np.array(one2oneAA) >= th)
+        FP_ = sum(np.array(one2oneAB) >= th)
+        FN_ = sum(np.array(one2oneAA) < th)
+        TN_ = sum(np.array(one2oneAB) < th)
+        PPrecise_.append(TP_/(TP_+FP_+1e-6))
+        PRecall_.append(TP_/(TP_+FN_+1e-6))
+        NPrecise_.append(TN_/(TN_+FN_+1e-6))
+        NRecall_.append(TN_/(TN_+FP_+1e-6))
+        
+        '''============================= 1:n'''
+        TP = sum(np.array(tpsimi) >= th)
+        FP = sum(np.array(fpsimi) >= th)
+        FN = sum(np.array(fnsimi) < th)
+        TN = sum(np.array(tnsimi) < th)
+        PPrecise.append(TP/(TP+FP+1e-6))
+        PRecall.append(TP/(TP+FN+1e-6))
+        NPrecise.append(TN/(TN+FN+1e-6))
+        NRecall.append(TN/(TN+FP+1e-6))
+        
+        
+        '''============================= 1:N 展厅'''
+        TPX = sum(np.array(tp_simi) >= th)
+        FPX = sum(np.array(fp_simi) >= th)
+        FNX = sum(np.array(fn_simi) < th)
+        TNX = sum(np.array(tn_simi) < th)
+        PPreciseX.append(TPX/(TPX+FPX+1e-6))
+        PRecallX.append(TPX/(TPX+FNX+1e-6))
+        NPreciseX.append(TNX/(TNX+FNX+1e-6))
+        NRecallX.append(TNX/(TNX+FPX+1e-6))
+    
+    '''============================= 1:1 曲线'''
+    fig, ax = plt.subplots()
+    ax.plot(Thresh, PPrecise_, 'r', label='Precise_Pos: TP/TPFP')
+    ax.plot(Thresh, PRecall_, 'b', label='Recall_Pos: TP/TPFN')
+    ax.plot(Thresh, NPrecise_, 'g', label='Precise_Neg: TN/TNFP')
+    ax.plot(Thresh, NRecall_, 'c', label='Recall_Neg: TN/TNFN')
+    ax.set_xlim([0, 1])
+    ax.set_ylim([0, 1])
+    ax.grid(True)
+    ax.set_title('Precise & Recall')
+    ax.set_xlabel(f"Num: {len(evtpaths)}")     
+    ax.legend()
+    plt.show()
+    
+    '''============================= 1:1 直方图'''
+    fig, axes = plt.subplots(2, 1)
+    axes[0].hist(np.array(one2oneAA), bins=60, edgecolor='black')
+    axes[0].set_xlim([-0.2, 1])
+    axes[0].set_title('AA')
+    axes[1].hist(np.array(one2oneAB), bins=60, edgecolor='black')
+    axes[1].set_xlim([-0.2, 1])
+    axes[1].set_title('BB')
+    plt.show()
+        
+    '''============================= 1:n 曲线''' 
+    fig, ax = plt.subplots()
+    ax.plot(Thresh, PPrecise, 'r', label='Precise_Pos: TP/TPFP')
+    ax.plot(Thresh, PRecall, 'b', label='Recall_Pos: TP/TPFN')
+    ax.plot(Thresh, NPrecise, 'g', label='Precise_Neg: TN/TNFP')
+    ax.plot(Thresh, NRecall, 'c', label='Recall_Neg: TN/TNFN')
+    ax.set_xlim([0, 1])
+    ax.set_ylim([0, 1])
+    ax.grid(True)
+    ax.set_title('Precise & Recall')
+    ax.set_xlabel(f"Num: {len(evtpaths)}")     
+    ax.legend()
+    plt.show()
+    
+    '''============================= 1:n 直方图'''
+    fig, axes = plt.subplots(2, 2)
+    axes[0, 0].hist(tpsimi, bins=60, edgecolor='black')
+    axes[0, 0].set_xlim([-0.2, 1])
+    axes[0, 0].set_title('TP')
+    axes[0, 1].hist(fpsimi, bins=60, edgecolor='black')
+    axes[0, 1].set_xlim([-0.2, 1])
+    axes[0, 1].set_title('FP')
+    axes[1, 0].hist(tnsimi, bins=60, edgecolor='black')
+    axes[1, 0].set_xlim([-0.2, 1])
+    axes[1, 0].set_title('TN')
+    axes[1, 1].hist(fnsimi, bins=60, edgecolor='black')
+    axes[1, 1].set_xlim([-0.2, 1])
+    axes[1, 1].set_title('FN')
+    plt.show()
+    
+    
+    '''============================= 1:N 展厅 曲线''' 
+    fig, ax = plt.subplots()
+    ax.plot(Thresh, PPreciseX, 'r', label='Precise_Pos: TP/TPFP')
+    ax.plot(Thresh, PRecallX, 'b', label='Recall_Pos: TP/TPFN')
+    ax.plot(Thresh, NPreciseX, 'g', label='Precise_Neg: TN/TNFP')
+    ax.plot(Thresh, NRecallX, 'c', label='Recall_Neg: TN/TNFN')
+    ax.set_xlim([0, 1])
+    ax.set_ylim([0, 1])
+    ax.grid(True)
+    ax.set_title('Precise & Recall')
+    ax.set_xlabel(f"Num: {len(evtpaths)}")     
+    ax.legend()
+    plt.show()
+    
+    '''============================= 1:N 展厅 直方图'''
+    fig, axes = plt.subplots(2, 2)
+    axes[0, 0].hist(tp_simi, bins=60, edgecolor='black')
+    axes[0, 0].set_xlim([-0.2, 1])
+    axes[0, 0].set_title('TP')
+    axes[0, 1].hist(fp_simi, bins=60, edgecolor='black')
+    axes[0, 1].set_xlim([-0.2, 1])
+    axes[0, 1].set_title('FP')
+    axes[1, 0].hist(tn_simi, bins=60, edgecolor='black')
+    axes[1, 0].set_xlim([-0.2, 1])
+    axes[1, 0].set_title('TN')
+    axes[1, 1].hist(fn_simi, bins=60, edgecolor='black')
+    axes[1, 1].set_xlim([-0.2, 1])
+    axes[1, 1].set_title('FN')
+    plt.show()
+    
+    
+
+    
+    print('Done!')
+            
+
     
 
 if __name__ == "__main__":
-    main()
+    evtpaths = r"\\192.168.1.28\share\测试视频数据以及日志\各模块测试记录\展厅测试\1120_展厅模型v801测试\扫A放A"
+    one2one_pr(evtpaths)