YOLOv5 分類模型 數(shù)據(jù)集加載 3 自定義類別

flyfish

YOLOv5 分類模型 數(shù)據(jù)集加載 1 樣本處理
YOLOv5 分類模型 數(shù)據(jù)集加載 2 切片處理
YOLOv5 分類模型的預(yù)處理（1） Resize 和 CenterCrop
YOLOv5 分類模型的預(yù)處理（2）ToTensor 和 Normalize
YOLOv5 分類模型 Top 1和Top 5 指標(biāo)說明
YOLOv5 分類模型 Top 1和Top 5 指標(biāo)實(shí)現(xiàn)

之前的處理方式是類別名字是文件夾名字，類別ID是按照文件夾名字的字母順序
現(xiàn)在是類別名字是文件夾名字，按照文件列表名字順序 例如

classes_name=['n02086240', 'n02087394', 'n02088364', 'n02089973', 'n02093754', 
'n02096294', 'n02099601', 'n02105641', 'n02111889', 'n02115641']

n02086240 類別ID是0
n02087394 類別ID是1
代碼處理是

if classes_name is None or not classes_name:classes, class_to_idx = self.find_classes(self.root)print("not classes_name")else:classes = classes_nameclass_to_idx ={cls_name: i for i, cls_name in enumerate(classes)}print("is classes_name")

完整

import time
from models.common import DetectMultiBackend
import os
import os.path
from typing import Any, Callable, cast, Dict, List, Optional, Tuple, Union
import cv2
import numpy as npimport torch
from PIL import Image
import torchvision.transforms as transformsimport sysclasses_name=['n02086240', 'n02087394', 'n02088364', 'n02089973', 'n02093754', 'n02096294', 'n02099601', 'n02105641', 'n02111889', 'n02115641']class DatasetFolder:def __init__(self,root: str,) -> None:self.root = rootif classes_name is None or not classes_name:classes, class_to_idx = self.find_classes(self.root)print("not classes_name")else:classes = classes_nameclass_to_idx ={cls_name: i for i, cls_name in enumerate(classes)}print("is classes_name")print("classes:",classes)print("class_to_idx:",class_to_idx)samples = self.make_dataset(self.root, class_to_idx)self.classes = classesself.class_to_idx = class_to_idxself.samples = samplesself.targets = [s[1] for s in samples]@staticmethoddef make_dataset(directory: str,class_to_idx: Optional[Dict[str, int]] = None,) -> List[Tuple[str, int]]:directory = os.path.expanduser(directory)if class_to_idx is None:_, class_to_idx = self.find_classes(directory)elif not class_to_idx:raise ValueError("'class_to_index' must have at least one entry to collect any samples.")instances = []available_classes = set()for target_class in sorted(class_to_idx.keys()):class_index = class_to_idx[target_class]target_dir = os.path.join(directory, target_class)if not os.path.isdir(target_dir):continuefor root, _, fnames in sorted(os.walk(target_dir, followlinks=True)):for fname in sorted(fnames):path = os.path.join(root, fname)if 1:  # 驗(yàn)證:item = path, class_indexinstances.append(item)if target_class not in available_classes:available_classes.add(target_class)empty_classes = set(class_to_idx.keys()) - available_classesif empty_classes:msg = f"Found no valid file for the classes {', '.join(sorted(empty_classes))}. "return instancesdef find_classes(self, directory: str) -> Tuple[List[str], Dict[str, int]]:classes = sorted(entry.name for entry in os.scandir(directory) if entry.is_dir())if not classes:raise FileNotFoundError(f"Couldn't find any class folder in {directory}.")class_to_idx = {cls_name: i for i, cls_name in enumerate(classes)}return classes, class_to_idxdef __getitem__(self, index: int) -> Tuple[Any, Any]:path, target = self.samples[index]sample = self.loader(path)return sample, targetdef __len__(self) -> int:return len(self.samples)def loader(self, path):print("path:", path)#img = cv2.imread(path)  # BGR HWCimg=Image.open(path).convert("RGB") # RGB HWCreturn imgdef time_sync():return time.time()#sys.exit() 
dataset = DatasetFolder(root="/media/a/flyfish/source/yolov5/datasets/imagewoof/val")#image, label=dataset[7]#
weights = "/home/a/classes.pt"
device = "cpu"
model = DetectMultiBackend(weights, device=device, dnn=False, fp16=False)
model.eval()
print(model.names)
print(type(model.names))mean=[0.485, 0.456, 0.406]
std=[0.229, 0.224, 0.225]
def preprocess(images):#實(shí)現(xiàn) PyTorch Resizetarget_size =224img_w = images.widthimg_h = images.heightif(img_h >= img_w):# hwresize_img = images.resize((target_size, int(target_size * img_h / img_w)), Image.BILINEAR)else:resize_img = images.resize((int(target_size * img_w  / img_h),target_size), Image.BILINEAR)#實(shí)現(xiàn) PyTorch CenterCropwidth = resize_img.widthheight = resize_img.heightcenter_x,center_y = width//2,height//2left = center_x - (target_size//2)top = center_y- (target_size//2)right =center_x +target_size//2bottom = center_y+target_size//2cropped_img = resize_img.crop((left, top, right, bottom))#實(shí)現(xiàn) PyTorch ToTensor Normalizeimages = np.asarray(cropped_img)print("preprocess:",images.shape)images = images.astype('float32')images = (images/255-mean)/stdimages = images.transpose((2, 0, 1))# HWC to CHWprint("preprocess:",images.shape)images = np.ascontiguousarray(images)images=torch.from_numpy(images)#images = images.unsqueeze(dim=0).float()return imagespred, targets, loss, dt = [], [], 0, [0.0, 0.0, 0.0]
# current batch size =1
for i, (images, labels) in enumerate(dataset):print("i:", i)im = preprocess(images)images = im.unsqueeze(0).to("cpu").float()print(images.shape)t1 = time_sync()images = images.to(device, non_blocking=True)t2 = time_sync()# dt[0] += t2 - t1y = model(images)y=y.numpy()#print("y:", y)t3 = time_sync()# dt[1] += t3 - t2#batch size >1 圖像推理結(jié)果是二維的#y: [[     4.0855     -1.1707     -1.4998      -0.935     -1.9979      -2.258     -1.4691     -1.0867     -1.9042    -0.99979]]tmp1=y.argsort()[:,::-1][:, :5]#batch size =1 圖像推理結(jié)果是一維的， 就要處理下argsort的維度#y: [     3.7441      -1.135     -1.1293     -0.9422     -1.6029     -2.0561      -1.025     -1.5842     -1.3952     -1.1824]#print("tmp1:", tmp1)pred.append(tmp1)#print("labels:", labels)targets.append(labels)#print("for pred:", pred)  # list#print("for targets:", targets)  # list# dt[2] += time_sync() - t3pred, targets = np.concatenate(pred), np.array(targets)
print("pred:", pred)
print("pred:", pred.shape)
print("targets:", targets)
print("targets:", targets.shape)
correct = ((targets[:, None] == pred)).astype(np.float32)
print("correct:", correct.shape)
print("correct:", correct)
acc = np.stack((correct[:, 0], correct.max(1)), axis=1)  # (top1, top5) accuracy
print("acc:", acc.shape)
print("acc:", acc)
top = acc.mean(0)
print("top1:", top[0])
print("top5:", top[1])