提问人：Xing 提问时间：11/17/2023 最后编辑：RandallXing 更新时间：11/18/2023 访问量：24
尝试使用 pytorch 编写 yolo1，但在训练时得到 NaN
Trying to write yolo1 with pytorch, but get NaN when train
问：
拜托，我想知道我的模型或损失函数或数据集是否错误。
我尝试打印了很多东西进行调试，但仍然不知道问题出在哪里。
我的源代码是：
import time
import torch
import torch.nn as nn
import cv2
import os
import numpy as np
from PIL import Image
from tqdm import tqdm
from torchvision import transforms
import torchvision
import torch.optim.lr_scheduler as lr_scheduler
NC = 102 # 类别数
BC = 2 # 锚框数
CLASSES = []
transform = transforms.Compose([
transforms.Resize(448, antialias=True), # 缩放
transforms.RandomCrop(448), # 随机裁剪
transforms.RandomHorizontalFlip(), # 水平翻转
transforms.RandomVerticalFlip(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]), # 标准化
])
class Dataset(torch.utils.data.Dataset):
def __init__(self, data_path, label_path):
self.data_path = data_path
self.label_path = label_path
self.data = []
for i in tqdm(os.listdir(data_path)):
img = os.path.join(data_path, i)
label = os.path.join(label_path, i.split('.')[0] + '.txt')
self.data.append([img, label])
def __getitem__(self, index):
img = cv2.imread(self.data[index][0])
# img = Image.open(self.data[index][0]).convert('RGB')
h, w = img.shape[0], img.shape[1]
padh, padw = 0, 0
# if h > w:
# padw = (h - w) // 2
# img = np.pad(img, ((0, 0), (padw, h - w - padw), (0, 0)), 'constant', constant_values=0)
# elif h < w:
# padh = (w - h) // 2
# img = np.pad(img, ((padh, w - h - padh), (0, 0), (0, 0)), 'constant', constant_values=0)
img = cv2.resize(img, (448, 448))
img = torch.from_numpy(np.array(img)).permute(2, 0, 1).float() / 255.0
label = open(self.data[index][1]).readlines()
label = [[float(j) for j in i.strip().split()] for i in label]
for i in range(len(label)):
if len(label[i]) % 5 != 0:
raise ValueError('label error')
if padw != 0:
label[i][2] = (label[i][2] * h + padh) / h
label[i][3] = (label[i][3] * w + h - w) / h
if padh != 0:
label[i][1] = (label[i][1] * w + padw) / w
label[i][4] = (label[i][4] * h + w - h) / w
# label = [x for y in label for x in y] # 把label拉成一个列表
# 进行了左侧和上边的填充， 改变图片大小，那么原来的框位置就出现了偏移，也需要相应修改
# print(label)
label = deal_label(label)
label = torch.tensor(label).float()
return img, label, self.data[index][0]
def __len__(self):
return len(self.data)
def deal_label(label_list):
# [ # label_list
# [label,x, y, w, h]
# ]
gridsize = 1.0 / 7 # 7最终划分网格数
labels = np.zeros((7, 7, (NC + 5 * BC))) # （x网格数，y网格数，5*锚框数+类别数）
for i in label_list:
gridx = int(i[1] // gridsize)
gridy = int(i[2] // gridsize)
labels[gridx, gridy, 5 * 2 + int(i[0])] = 1
labels[gridx, gridy, 0:5] = np.array([1, i[1], i[2], i[3], i[4]]) # confi,x,y,w,h
labels[gridx, gridy, 5:10] = np.array([1, i[1], i[2], i[3], i[4]])
return labels
data = Dataset(r'C:\Users\xingshuyin\python\datasets\disease_insect_detect\images\train',
r'C:\Users\xingshuyin\python\datasets\disease_insect_detect\labels\train')
def collate_fn(betch):
data = [item[0] for item in betch]
label = [item[1] for item in betch]
files = [item[2] for item in betch]
return torch.stack(data), torch.stack(label), files
loader = torch.utils.data.DataLoader(data, batch_size=8, shuffle=True, num_workers=2, collate_fn=collate_fn)
loader_val = torch.utils.data.DataLoader(data, batch_size=1, shuffle=True, num_workers=1, collate_fn=collate_fn)
def iou(b1, b2):
"""
b1 -> batch S S (NC + 5 * BC)
(NC + 5 * BC) -> confi cx cy w h confi cx cy w h cls1 cls2 ...
b2 -> batch S S (NC + 5 * BC)
"""
b11 = b1[0:5]
b12 = b1[5:10]
b21 = b2[0:5]
b22 = b2[5:10]
b11_area = b11[3] * b11[4]
b12_area = b12[3] * b12[4]
b21_area = b21[3] * b21[4]
b22_area = b22[3] * b22[4]
intersect_x_b11_b21 = torch.min(b11[1] + b11[3] / 2, b21[1] + b21[3] / 2) - torch.max(b11[1] - b11[3] / 2, b21[1] - b21[3] / 2)
intersect_y_b11_b21 = torch.min(b11[2] + b11[4] / 2, b21[2] + b21[4] / 2) - torch.max(b11[2] - b11[4] / 2, b21[2] - b21[4] / 2)
intersect_area_b11_b21 = intersect_x_b11_b21 * intersect_y_b11_b21
intersect_x_b12_b22 = torch.min(b12[1] + b12[3] / 2, b22[1] + b22[3] / 2) - torch.max(b12[1] - b12[3] / 2, b22[1] - b22[3] / 2)
intersect_y_b12_b22 = torch.min(b12[2] + b12[4] / 2, b22[2] + b22[4] / 2) - torch.max(b12[2] - b12[4] / 2, b22[2] - b22[4] / 2)
intersect_area_b12_b22 = intersect_x_b12_b22 * intersect_y_b12_b22
iou1 = intersect_area_b11_b21 / (b11_area + b21_area - intersect_area_b11_b21)
iou2 = intersect_area_b12_b22 / (b12_area + b22_area - intersect_area_b12_b22)
# print('b11 ', b11, 'b11_area ', b11_area)
# print('b12 ', b12, 'b12_area ', b12_area)
# print('b21 ', b21, 'b21_area ', b21_area)
# print('b22 ', b22, 'b22_area ', b22_area)
print('iou1 ', iou1, 'iou2 ', iou2)
return iou1 if iou1 > 0 else 0, iou2 if iou2 > 0 else 0
class YOLOLOSS(nn.Module):
def __init__(self, S):
super(YOLOLOSS, self).__init__()
self.S = S # 栅格数量
self.weight_xy = 5.0 # 坐标损失权重
self.weight_wh = 5.0 # 大小损失权重
self.weight_noobj_confi = 0.5 # 不包含物体的置信度损失权重
self.weight_obj_confi = 1 # 包含物体的置信度损失权重
self.weight_class = 1
def forward(self, pred, label):
"""
pred -> batch S S (NC + 5 * BC)
(NC + 5 * BC) -> confi cx cy w h confi cx cy w h cls1 cls2 ...
labal -> batch S S (NC + 5 * BC)
"""
batch_size = pred.size(0)
loss_xy_obj = 0
loss_wh_obj = 0
loss_obj_confi = 0
loss_noobj_confi = 0
loss_class = 0
for i in range(batch_size):
for x in range(self.S):
for y in range(self.S):
if label[i, x, y, 0] == 1 and label[i, x, y, 5] == 1: # 如果label在这个各自有值，这个各自就负责预测这个格子的label
iou1, iou2 = iou(pred[i, x, y], label[i, x, y])
if iou1 >= iou2:
loss_xy_obj += self.weight_xy * torch.sum((pred[i, x, y, 1:3] - label[i, x, y, 1:3])**2) # 公式第一行
loss_wh_obj += self.weight_wh * torch.sum((pred[i, x, y, 3:5].sqrt() - label[i, x, y, 3:5].sqrt())**2) # 公式第二行
loss_obj_confi += self.weight_obj_confi * torch.sum((pred[i, x, y, 0] - iou1)**2) # 公式第三行
loss_noobj_confi += self.weight_noobj_confi * torch.sum((pred[i, x, y, 5] - iou2)**2) # 公式第四行
else:
loss_xy_obj += self.weight_xy * torch.sum((pred[i, x, y, 6:8] - label[i, x, y, 6:8])**2)
loss_wh_obj += self.weight_wh * torch.sum((pred[i, x, y, 8:10].sqrt() - label[i, x, y, 8:10].sqrt())**2)
loss_obj_confi += self.weight_obj_confi * torch.sum((pred[i, x, y, 5] - iou2)**2)
loss_noobj_confi += self.weight_noobj_confi * torch.sum((pred[i, x, y, 0] - iou1)**2)
loss_class += self.weight_class * torch.sum((pred[i, x, y, 10:] - label[i, x, y, 10:])**2) # 公式第5行
else:
loss_noobj_confi += self.weight_noobj_confi * (torch.sum(pred[i, x, y, [0, 5]]**2)) # 公式第四行
print('loss ------------', 'loss_xy_obj ', loss_xy_obj, 'loss_wh_obj ', loss_wh_obj, 'loss_obj_confi ',
loss_obj_confi, 'loss_noobj_confi ', loss_noobj_confi, 'loss_class ', loss_class)
return (loss_xy_obj + loss_wh_obj + loss_obj_confi + loss_noobj_confi + loss_class) / batch_size
class YOLO(torch.nn.Module):
def __init__(self):
super(YOLO, self).__init__()
self.net = nn.Sequential(
nn.Conv2d(3, 64, 7, 2, 3),
nn.MaxPool2d(2, 2),
nn.Conv2d(64, 192, 3, 1, 1),
nn.MaxPool2d(2, 2),
nn.Conv2d(192, 128, 1, 1, 0), # 论文里没有padding 因为都是same ，所以大小为3的卷积核就需要padding=1
nn.Conv2d(128, 256, 3, 1, 1),
nn.Conv2d(256, 256, 1, 1, 0),
nn.Conv2d(256, 512, 3, 1, 1),
nn.MaxPool2d(2, 2),
nn.Conv2d(512, 256, 1, 1, 0),
nn.Conv2d(256, 512, 3, 1, 1),
nn.Conv2d(512, 256, 1, 1, 0),
nn.Conv2d(256, 512, 3, 1, 1),
nn.Conv2d(512, 256, 1, 1, 0),
nn.Conv2d(256, 512, 3, 1, 1),
nn.Conv2d(512, 256, 1, 1, 0),
nn.Conv2d(256, 512, 3, 1, 1),
# nn.Conv2d(512, 512, 1),
nn.Conv2d(512, 1024, 3, 1, 1),
nn.MaxPool2d(2, 2),
nn.Conv2d(1024, 512, 1, 1, 0),
nn.Conv2d(512, 1024, 3, 1, 1),
nn.Conv2d(1024, 512, 1, 1, 0),
nn.Conv2d(512, 1024, 3, 1, 1),
nn.LeakyReLU(),
# nn.Conv2d(1024, 1024, 3),
nn.Conv2d(1024, 1024, 3, 2, 1),
nn.BatchNorm2d(1024),
nn.LeakyReLU(),
nn.Conv2d(1024, 1024, 3, 1, 1),
nn.LeakyReLU(),
nn.Conv2d(1024, 1024, 3, 1, 1),
nn.LeakyReLU(),
)
self.net2 = nn.Sequential(
nn.Linear(7 * 7 * 1024, 4096),
nn.LeakyReLU(),
nn.Linear(4096, 7 * 7 * (NC + 5 * BC)),
nn.Sigmoid()
)
def forward(self, x):
"""
x -> betch S S (NC + 5 * BC)
"""
# TODO
out = self.net(x)
out = self.net2(out.view(out.size(0), -1))
# out = out * 100
return out.view(-1, 7, 7, (NC + 5 * BC)) # 返回的数据包含betxh
def train():
m = YOLO() # 创建模型
m.to('cuda') # 把模型放到cuda上
loss = YOLOLOSS(7) # 创建损失函数
lr = 0.01
opt = torch.optim.SGD(m.parameters(), lr=lr, momentum=0.9, weight_decay=0.0005) # 创建优化器
# scheduler = lr_scheduler.CosineAnnealingWarmRestarts(opt, T_0=100, T_mult=5, eta_min=0.001)
loss_min = 9999
for E in range(5):
pbar = tqdm(loader)
for i in pbar:
m.zero_grad() # 梯度清零, 不然数据会累加
data, label, files = i[0].to('cuda'), i[1].to('cuda'), i[2]
print(data.size())
r = m(data)
# print(r.size())
# 神经网络可视化
# g = make_dot(r ,params=dict(m.named_parameters()), show_attrs=True, show_saved=True)
# g.render(filename='graph', view=False)
print(data[0][0][0])
print(label[0][0][0])
print(r[0][0][0])
l = loss(r, label) # 计算loss
pbar.set_description(F"epoch {E} | loss_min: {loss_min} | loss: {l} | lr:{lr} ")
# print(files)
# loss.grad_fn.next_functions[0][0].next_functions 可以逐级获取传播过程
if l < loss_min and l < 5:
loss_min = l
torch.save(m.state_dict(), './model')
l.backward() # 反向传播
opt.step() # 更新参数
# scheduler.step()
lr = opt.param_groups[0]['lr']
time.sleep(0.5)
def pred_to_box(pred):
boxs = []
for i in range(len(pred)):
b = []
for j in range(len(pred[i])):
b.append(torch.cat(pred[i][j][:5], pred[i][j][10:]))
b.append(torch.cat(pred[i][j][5:10], pred[i][j][10:]))
boxs.append(b)
return boxs
def draw_box(img, boxs):
...
def val():
m = YOLO()
m.load_state_dict(torch.load('./model'))
m.eval()
for i in tqdm(loader_val):
data, label, files = i[0].to('cuda'), i[1].to('cuda'), i[2]
r = m(data)
date = []
if __name__ == '__main__':
torch.set_printoptions(profile="full")
train()
运行程序时，我得到此输出，其中显示了第 3 个数据。nantensor
loss ------------ loss_xy_obj tensor(24.2689, device='cuda:0', grad_fn=<AddBackward0>) loss_wh_obj tensor(16.9818, device='cuda:0', grad_fn=<AddBackward0>) loss_obj_confi tensor(4.2992, device='cuda:0', grad_fn=<AddBackward0>) loss_noobj_confi tensor(3.0024e-05, device='cuda:0', grad_fn=<AddBackward0>) loss_class tensor(67.2501, device='cuda:0', grad_fn=<AddBackward0>)
epoch 0 | loss_min: 9999 | loss: 14.099996566772461 | lr:0.01 : 0%|▊ | 9/2024 [00:10<27:03, 1.24it/s]torch.Size([8, 3, 448, 448])
tensor([0.3020, 0.3020, 0.3020, 0.2980, 0.2941, 0.2941, 0.2941, 0.2941, 0.2941,
0.2902, 0.2863, 0.2824, 0.2784, 0.2745, 0.2784, 0.2784, 0.2784, 0.2745,
0.2667, 0.2627, 0.2627, 0.2549, 0.2549, 0.2588, 0.2588, 0.2627, 0.2667,
0.2706, 0.2667, 0.2549, 0.2627, 0.2706, 0.2706, 0.2667, 0.2588, 0.2549,
0.2706, 0.2706, 0.2706, 0.2706, 0.2706, 0.2706, 0.2706, 0.2706, 0.2706,
0.2706, 0.2706, 0.2706, 0.2706, 0.2706, 0.2745, 0.2745, 0.2745, 0.2745,
0.2745, 0.2745, 0.2745, 0.2706, 0.2706, 0.2706, 0.2745, 0.2745, 0.2745,
0.2784, 0.2863, 0.2902, 0.2863, 0.2824, 0.2863, 0.2902, 0.2980, 0.3020,
0.2980, 0.2980, 0.2980, 0.2980, 0.2980, 0.2980, 0.2980, 0.3059, 0.3059,
0.3059, 0.3098, 0.3137, 0.3137, 0.3137, 0.3059, 0.3020, 0.3020, 0.3020,
0.3020, 0.3098, 0.3137, 0.3020, 0.3059, 0.3098, 0.3137, 0.3137, 0.3137,
0.3098, 0.3098, 0.3098, 0.3098, 0.3098, 0.3098, 0.3098, 0.3098, 0.3098,
0.3098, 0.3098, 0.3098, 0.3098, 0.3098, 0.3098, 0.3137, 0.3216, 0.3216,
0.3216, 0.3216, 0.3216, 0.3216, 0.3216, 0.3216, 0.3216, 0.3216, 0.3216,
0.3216, 0.3216, 0.3216, 0.3255, 0.3216, 0.3216, 0.3176, 0.3176, 0.3216,
0.3255, 0.3176, 0.3176, 0.3176, 0.3176, 0.3176, 0.3176, 0.3176, 0.3176,
0.3176, 0.3176, 0.3176, 0.3176, 0.3176, 0.3176, 0.3176, 0.3176, 0.3176,
0.3137, 0.3137, 0.3176, 0.3176, 0.3216, 0.3216, 0.3216, 0.3176, 0.3176,
0.3137, 0.3137, 0.3137, 0.3137, 0.3137, 0.3137, 0.3137, 0.3137, 0.3137,
0.3137, 0.3059, 0.3059, 0.3098, 0.3137, 0.3137, 0.3216, 0.3216, 0.3137,
0.3137, 0.3137, 0.3137, 0.3137, 0.3137, 0.3137, 0.3137, 0.3137, 0.3137,
0.3137, 0.3137, 0.3137, 0.3137, 0.3137, 0.3137, 0.3137, 0.3098, 0.3098,
0.3137, 0.3137, 0.3137, 0.3098, 0.3098, 0.3098, 0.3098, 0.3098, 0.3098,
0.3098, 0.3176, 0.3137, 0.3098, 0.3098, 0.3098, 0.3137, 0.3176, 0.3137,
0.3137, 0.3137, 0.3137, 0.3137, 0.3137, 0.3137, 0.3098, 0.3098, 0.3098,
0.3098, 0.3098, 0.3098, 0.3098, 0.3176, 0.3176, 0.3098, 0.3098, 0.3098,
0.3098, 0.3176, 0.3137, 0.3059, 0.3059, 0.3020, 0.3020, 0.3059, 0.3059,
0.3098, 0.3059, 0.3059, 0.3059, 0.3059, 0.3059, 0.2980, 0.2980, 0.2941,
0.2941, 0.2902, 0.2863, 0.2784, 0.2784, 0.2784, 0.2980, 0.2980, 0.2980,
0.2941, 0.2902, 0.2902, 0.2902, 0.2784, 0.2824, 0.2824, 0.2863, 0.2863,
0.2902, 0.2902, 0.2902, 0.2902, 0.2902, 0.2902, 0.2902, 0.2902, 0.2902,
0.2902, 0.2902, 0.2902, 0.2902, 0.2902, 0.2902, 0.2902, 0.2902, 0.2902,
0.2824, 0.2745, 0.2745, 0.2784, 0.2863, 0.2902, 0.2784, 0.2784, 0.2745,
0.2745, 0.2706, 0.2706, 0.2706, 0.2627, 0.2667, 0.2745, 0.2784, 0.2745,
0.2706, 0.2627, 0.2627, 0.2627, 0.2627, 0.2627, 0.2627, 0.2627, 0.2667,
0.2784, 0.2863, 0.2863, 0.2863, 0.2824, 0.2980, 0.3725, 0.4431, 0.4980,
0.5412, 0.5608, 0.5608, 0.5529, 0.5608, 0.5569, 0.5373, 0.4941, 0.4275,
0.3490, 0.2706, 0.2078, 0.1725, 0.1922, 0.2078, 0.2196, 0.2157, 0.1922,
0.1608, 0.1412, 0.1255, 0.1333, 0.1412, 0.1490, 0.1529, 0.1490, 0.1451,
0.1373, 0.1294, 0.1255, 0.1216, 0.1137, 0.1255, 0.1333, 0.1569, 0.1765,
0.1882, 0.1922, 0.1804, 0.1686, 0.1569, 0.1569, 0.1647, 0.1686, 0.1647,
0.1529, 0.1490, 0.1529, 0.1686, 0.1647, 0.1647, 0.1608, 0.1647, 0.1686,
0.1804, 0.1961, 0.2510, 0.2667, 0.2863, 0.2941, 0.2941, 0.2784, 0.2667,
0.2627, 0.2588, 0.2588, 0.2471, 0.2353, 0.2196, 0.2078, 0.2078, 0.2078,
0.2078, 0.2235, 0.2431, 0.2627, 0.2667, 0.2627, 0.2588, 0.2627, 0.2706,
0.2824, 0.2706, 0.2588, 0.2627, 0.2627, 0.2431, 0.2314, 0.2353, 0.2275,
0.2235, 0.2039, 0.2039, 0.2196, 0.2314, 0.2314, 0.2314, 0.2353, 0.2431,
0.2667, 0.2588, 0.2588, 0.2627, 0.2667, 0.2627, 0.2510, 0.2196, 0.2235,
0.2549, 0.2784, 0.2863, 0.2902, 0.2824, 0.2824, 0.3176, 0.3412, 0.3294,
0.3216, 0.3333, 0.3333, 0.2941, 0.2510, 0.2275, 0.2039],
device='cuda:0')
tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],
device='cuda:0')
tensor([nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan],
device='cuda:0', grad_fn=<SelectBackward0>)
iou1 tensor(nan, device='cuda:0', grad_fn=<DivBackward0>) iou2 tensor(nan, device='cuda:0', grad_fn=<DivBackward0>)
iou1 tensor(nan, device='cuda:0', grad_fn=<DivBackward0>) iou2 tensor(nan, device='cuda:0', grad_fn=<DivBackward0>)
iou1 tensor(nan, device='cuda:0', grad_fn=<DivBackward0>) iou2 tensor(nan, device='cuda:0', grad_fn=<DivBackward0>)
iou1 tensor(nan, device='cuda:0', grad_fn=<DivBackward0>) iou2 tensor(nan, device='cuda:0', grad_fn=<DivBackward0>)
iou1 tensor(nan, device='cuda:0', grad_fn=<DivBackward0>) iou2 tensor(nan, device='cuda:0', grad_fn=<DivBackward0>)
iou1 tensor(nan, device='cuda:0', grad_fn=<DivBackward0>) iou2 tensor(nan, device='cuda:0', grad_fn=<DivBackward0>)
iou1 tensor(nan, device='cuda:0', grad_fn=<DivBackward0>) iou2 tensor(nan, device='cuda:0', grad_fn=<DivBackward0>)
iou1 tensor(nan, device='cuda:0', grad_fn=<DivBackward0>) iou2 tensor(nan, device='cuda:0', grad_fn=<DivBackward0>)
loss ------------ loss_xy_obj tensor(nan, device='cuda:0', grad_fn=<AddBackward0>) loss_wh_obj tensor(nan, device='cuda:0', grad_fn=<AddBackward0>) loss_obj_confi tensor(nan, device='cuda:0', grad_fn=<AddBackward0>) loss_noobj_confi tensor(nan, device='cuda:0', grad_fn=<AddBackward0>) loss_class tensor(nan, device='cuda:0', grad_fn=<AddBackward0>)
epoch 0 | loss_min: 9999 | loss: nan | lr:0.01 : 0%|▉ | 10/2024 [00:11<26:03, 1.29it/s]torch.Size([8, 3, 448, 448])
tensor([0.6549, 0.7216, 0.7098, 0.4941, 0.3294, 0.2745, 0.3529, 0.3843, 0.4275,
0.4941, 0.6275, 0.6863, 0.7137, 0.7529, 0.7137, 0.6353, 0.5216, 0.4941,
0.4863, 0.4980, 0.5020, 0.5137, 0.5373, 0.5294, 0.5333, 0.5451, 0.5569,
0.5569, 0.5569, 0.5647, 0.5765, 0.5882, 0.6000, 0.6078, 0.6196, 0.6353,
0.6392, 0.6392, 0.6392, 0.6471, 0.6549, 0.6588, 0.6627, 0.6627, 0.6667,
0.6667, 0.6667, 0.6667, 0.6706, 0.6824, 0.6902, 0.6941, 0.7020, 0.7059,
0.7137, 0.7216, 0.7216, 0.7176, 0.7137, 0.7059, 0.7020, 0.6980, 0.6784,
0.6706, 0.6706, 0.6667, 0.6627, 0.6588, 0.6588, 0.6510, 0.6392, 0.6353,
0.6314, 0.6314, 0.6314, 0.6314, 0.6275, 0.6196, 0.6118, 0.6039, 0.5961,
0.6039, 0.6039, 0.6000, 0.5961, 0.6000, 0.6000, 0.6000, 0.5882, 0.5725,
0.5490, 0.5294, 0.5333, 0.5412, 0.5373, 0.5333, 0.5294, 0.5294, 0.5333,
0.5373, 0.5451, 0.5569, 0.5647, 0.5686, 0.5725, 0.5725, 0.5804, 0.5882,
0.5961, 0.6078, 0.6196, 0.6314, 0.6431, 0.6549, 0.6667, 0.6745, 0.6784,
0.6902, 0.6980, 0.7059, 0.7059, 0.7020, 0.7020, 0.7020, 0.7059, 0.7059,
0.7098, 0.6941, 0.6902, 0.6902, 0.6902, 0.6902, 0.6902, 0.6902, 0.6902,
0.6902, 0.6863, 0.6784, 0.6745, 0.6745, 0.6667, 0.6667, 0.6667, 0.6706,
0.6706, 0.6745, 0.6784, 0.6824, 0.6863, 0.6902, 0.6980, 0.7020, 0.7020,
0.7020, 0.7020, 0.7020, 0.6980, 0.6941, 0.6941, 0.6902, 0.6863, 0.6784,
0.6745, 0.6706, 0.6706, 0.6706, 0.6667, 0.6627, 0.6627, 0.6588, 0.6510,
0.6431, 0.6392, 0.6275, 0.6157, 0.6157, 0.6157, 0.6196, 0.6235, 0.6157,
0.6157, 0.6157, 0.6118, 0.6118, 0.6118, 0.6078, 0.6078, 0.6118, 0.6118,
0.6078, 0.6039, 0.6000, 0.5922, 0.5961, 0.6039, 0.6078, 0.6039, 0.6000,
0.6078, 0.6157, 0.6235, 0.6275, 0.6314, 0.6314, 0.6353, 0.6510, 0.6627,
0.6667, 0.6667, 0.6706, 0.6745, 0.6824, 0.6902, 0.7020, 0.7098, 0.7137,
0.7176, 0.7176, 0.7176, 0.7294, 0.7490, 0.7333, 0.7294, 0.7333, 0.7412,
0.7333, 0.7333, 0.7412, 0.7490, 0.7490, 0.7490, 0.7725, 0.7843, 0.7804,
0.7725, 0.7725, 0.7765, 0.7843, 0.7843, 0.7804, 0.7725, 0.7725, 0.7725,
0.7765, 0.7725, 0.7725, 0.7686, 0.7725, 0.7725, 0.7765, 0.7804, 0.7765,
0.7765, 0.7804, 0.7804, 0.7804, 0.7765, 0.7765, 0.7765, 0.7725, 0.7647,
0.7569, 0.7569, 0.7529, 0.7490, 0.7373, 0.6941, 0.7137, 0.7216, 0.7020,
0.7294, 0.7176, 0.6667, 0.6196, 0.6471, 0.6980, 0.6784, 0.6353, 0.5922,
0.5843, 0.5686, 0.5451, 0.5137, 0.4902, 0.4667, 0.4510, 0.4431, 0.4431,
0.4353, 0.4196, 0.4118, 0.4078, 0.4118, 0.4118, 0.4118, 0.4196, 0.4196,
0.4235, 0.4314, 0.4235, 0.4235, 0.4275, 0.4235, 0.4235, 0.4275, 0.4353,
0.4392, 0.4431, 0.4510, 0.4510, 0.4549, 0.4588, 0.4745, 0.4824, 0.4863,
0.4980, 0.5020, 0.5020, 0.5059, 0.5059, 0.5098, 0.5216, 0.5255, 0.5216,
0.5098, 0.5098, 0.5098, 0.5098, 0.5059, 0.4980, 0.4902, 0.4784, 0.4706,
0.4588, 0.4510, 0.4314, 0.4196, 0.4118, 0.3961, 0.3882, 0.3804, 0.3725,
0.3686, 0.3608, 0.3529, 0.3490, 0.3412, 0.3294, 0.3373, 0.3490, 0.3608,
0.3765, 0.3882, 0.4000, 0.4196, 0.4353, 0.4510, 0.4627, 0.4706, 0.4784,
0.4980, 0.5059, 0.5098, 0.5098, 0.5098, 0.5098, 0.5137, 0.5059, 0.4902,
0.4824, 0.4941, 0.4824, 0.4627, 0.4353, 0.4157, 0.4000, 0.3922, 0.3765,
0.3608, 0.3412, 0.3333, 0.3255, 0.3176, 0.3098, 0.3020, 0.2902, 0.2863,
0.2863, 0.2902, 0.2902, 0.2863, 0.2902, 0.2980, 0.3137, 0.3490, 0.3882,
0.4314, 0.4745, 0.5176, 0.5569, 0.5961, 0.6314, 0.6667, 0.6902, 0.7098,
0.7255, 0.7529, 0.7765, 0.7922, 0.8000, 0.8039, 0.8078, 0.8039, 0.8039,
0.7961, 0.7804, 0.7608, 0.7490, 0.7412, 0.7333, 0.7216, 0.7137, 0.7059,
0.6941, 0.6784, 0.6706, 0.6588, 0.6510, 0.6549, 0.6431, 0.6353, 0.6275,
0.6157, 0.6157, 0.6157, 0.6078, 0.6118, 0.6157, 0.6235, 0.6314, 0.6431,
0.6510, 0.6510, 0.6510, 0.6549, 0.6549, 0.6510, 0.6510],
device='cuda:0')
tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],
device='cuda:0')
tensor([nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan],
device='cuda:0', grad_fn=<SelectBackward0>)
iou1 tensor(nan, device='cuda:0', grad_fn=<DivBackward0>) iou2 tensor(nan, device='cuda:0', grad_fn=<DivBackward0>)
Exception ignored in: <function _MultiProcessingDataLoaderIter.__del__ at 0x000001CE90375630>
Traceback (most recent call last):
File "C:\Users\xingshuyin\app\miniconda\envs\yolov5\lib\site-packages\torch\utils\data\dataloader.py", line 1478, in __del__
self._shutdown_workers()
File "C:\Users\xingshuyin\app\miniconda\envs\yolov5\lib\site-packages\torch\utils\data\dataloader.py", line 1442, in _shutdown_workers
w.join(timeout=_utils.MP_STATUS_CHECK_INTERVAL)
File "C:\Users\xingshuyin\app\miniconda\envs\yolov5\lib\multiprocessing\process.py", line 149, in join
res = self._popen.wait(timeout)
File "C:\Users\xingshuyin\app\miniconda\envs\yolov5\lib\multiprocessing\popen_spawn_win32.py", line 108, in wait
res = _winapi.WaitForSingleObject(int(self._handle), msecs)
KeyboardInterrupt:
epoch 0 | loss_min: 9999 | loss: nan | lr:0.01 : 0%|▉ | 10/2024 [00:12<40:23, 1.20s/it]
Traceback (most recent call last):
File "c:\Users\xingshuyin\python\yolov8\yolo1.py", line 354, in <module>
File "c:\Users\xingshuyin\python\yolov8\yolo1.py", line 311, in train
File "C:\Users\xingshuyin\app\miniconda\envs\yolov5\lib\site-packages\torch\nn\modules\module.py", line 1518, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "C:\Users\xingshuyin\app\miniconda\envs\yolov5\lib\site-packages\torch\nn\modules\module.py", line 1527, in _call_impl
return forward_call(*args, **kwargs)
File "c:\Users\xingshuyin\python\yolov8\yolo1.py", line 167, in forward
iou1, iou2 = iou(pred[i, x, y], label[i, x, y])
File "c:\Users\xingshuyin\python\yolov8\yolo1.py", line 135, in iou
print('iou1 ', iou1, 'iou2 ', iou2)
KeyboardInterrupt
答： 暂无答案
评论
draw_box