- 下载SSD代码：
- 将下载的代码解压后在data文件夹下新建VOCdevkit文件夹，然后将VOC2007数据集复制到该文
- 下载权重文件放在weights文件夹下面。下载地址：
- 修改config.py代码
1.# SSD300 CONFIGS
voc = {
'num_classes': 3,//将类别改为：类别+1（背景）
'lr_steps': (80000, 100000, 120000),
'max_iter': 120000,//迭代次数
'feature_maps': [38, 19, 10, 5, 3, 1],
'min_dim': 300,
'steps': [8, 16, 32, 64, 100, 300],
'min_sizes': [30, 60, 111, 162, 213, 264],
'max_sizes': [60, 111, 162, 213, 264, 315],
'aspect_ratios': [[2], [2, 3], [2, 3], [2, 3], [2], [2]],
'variance': [0.1, 0.2],
'clip': True,
'name': 'VOC',
} 修改VOC0712.py代码。
'''
VOC_CLASSES = ( # always index 0
'aeroplane', 'bicycle', 'bird', 'boat',
'bottle', 'bus', 'car', 'cat', 'chair',
'cow', 'diningtable', 'dog', 'horse',
'motorbike', 'person', 'pottedplant',
'sheep', 'sofa', 'train', 'tvmonitor')
'''
VOC_CLASSES = ( # always index 0
'aircraft', 'oiltank')//修改为自己数据集的类别。
# note: if you used our download scripts, this should be right
#VOC_ROOT = osp.join("", "data/VOCdevkit/")
VOC_ROOT = "data/VOCdevkit/" #修改为Win10路径 将
def __init__(self, root,
image_sets=[('2007', 'trainval'), ('2012', 'trainval')],
transform=None, target_transform=VOCAnnotationTransform(),
dataset_name='VOC0712'): 修改为：
def __init__(self, root,
image_sets=[('2007', 'trainval')],
transform=None, target_transform=VOCAnnotationTransform(),
dataset_name='VOC2007'): 修改coco.py
COCO_ROOT ='data/'
修改ssd.py
将32行的self.cfg = (coco, voc)[num_classes == 21]
修改为self.cfg = (coco, voc)[num_classes == 3]#3是我数据集的类别+1
修改multibox_loss.py
在97行的loss_c[pos] = 0 # filter out pos boxes for now
位置之上加入loss_c = loss_c.view(pos.size()[0], pos.size()[1])
如果没有这句话会引起张量不匹配的问题
如：
IndexError: The shape of the mask [4, 8732] at index 0 does not match the shape of the indexed tensor [34928, 1] at index 0
修改全局配置参数：
parser.add_argument('--dataset', default='VOC', choices=['VOC', 'COCO'],
type=str, help='VOC or COCO')#改为VOC
parser.add_argument('--batch_size', default=4, type=int,
help='Batch size for training')#Batch Size 按照显存的大小设置合适的值。
parser.add_argument('--start_iter', default=0, type=int,
help='Resume training at this iter')#迭代开始
parser.add_argument('--num_workers', default=4, type=int,
help='Number of workers used in dataloading')
parser.add_argument('--cuda', default=True, type=str2bool,
help='Use CUDA to train model')
parser.add_argument('--lr', '--learning-rate', default=1e-4, type=float,
help='initial learning rate')
parser.add_argument('--momentum', default=0.9, type=float,
help='Momentum value for optim')
parser.add_argument('--weight_decay', default=5e-4, type=float,
help='Weight decay for SGD')
parser.add_argument('--gamma', default=0.1, type=float,
help='Gamma update for SGD')
parser.add_argument('--visdom', default=False, type=str2bool,
help='Use visdom for loss visualization')
parser.add_argument('--save_folder', default='weights/',
help='Directory for saving checkpoint models')#保存权重的位置。
args = parser.parse_args()
将：
————————————————
loc_loss += loss_l.item()
conf_loss += loss_c.item()
if iteration % 10 == 0:
print('timer: %.4f sec.' % (t1 - t0))
print('iter ' + repr(iteration) + ' || Loss: %.4f ||' % (loss.item()), end=' ') 不修改会引起错误：
.IndexError: invalid index of a 0-dim tensor. Use tensor.item() to convert a 0-dim tensor to a Python number
将165行的
images, targets = next(batch_iterator)
更改成：
try:
images, targets = next(batch_iterator)
except StopIteration:
batch_iterator = iter(data_loader)
如果不改会引起跳出迭代的问题：
Traceback (most recent call last):
File "E:/ssd.pytorch-master/ssd.pytorch-master/train.py", line 246, in <module>
train()
File "E:/ssd.pytorch-master/ssd.pytorch-master/train.py", line 159, in train
images, targets = next(batch_iterator)
File "D:\Users\WH\anaconda3\lib\site-packages\torch\utils\data\dataloader.py", line 345, in __next__
data = self._next_data()
File "D:\Users\WH\anaconda3\lib\site-packages\torch\utils\data\dataloader.py", line 831, in _next_data
raise StopIteration
StopIteration
如果遇到loss为nan的现象，就需要降低出示学习率。
将10e-3改为10e-4
将以上问题解决后就可以开始训练了。
测试
需要修改test.py的全局参数。
parser.add_argument('--trained_model', default='weights/ssd300_COCO_60000.pth',
type=str, help='Trained state_dict file path to open')#指定测试用的模型。
parser.add_argument('--save_folder', default='eval/', type=str,
help='Dir to save results')#生成的测试结果
parser.add_argument('--visual_threshold', default=0.6, type=float,
help='Final confidence threshold')#检测的最低置信度。
点击测试生成测试结果：
GROUND TRUTH FOR: aircraft_27
label: 650.0 || 101.0 || 753.0 || 227.0 || 0
label: 395.0 || 351.0 || 521.0 || 464.0 || 0
label: 320.0 || 479.0 || 465.0 || 606.0 || 0
label: 276.0 || 617.0 || 432.0 || 753.0 || 0
PREDICTIONS:
1 label: aircraft score: tensor(1.0000) 317.00345 || 479.0164 || 468.34143 || 604.06415
2 label: aircraft score: tensor(1.0000) 279.88608 || 614.7385 || 434.32086 || 752.4983
3 label: aircraft score: tensor(0.9999) 397.0827 || 352.3487 || 524.27655 || 462.45944
4 label: aircraft score: tensor(0.9990) 647.89233 || 100.57587 || 758.556 || 232.56816
GROUND TRUTH FOR: oiltank_349
label: 115.0 || 340.0 || 237.0 || 466.0 || 1
label: 104.0 || 498.0 || 231.0 || 637.0 || 1
label: 92.0 || 675.0 || 224.0 || 814.0 || 1
label: 369.0 || 378.0 || 492.0 || 507.0 || 1
label: 357.0 || 543.0 || 483.0 || 678.0 || 1
label: 350.0 || 715.0 || 485.0 || 856.0 || 1
label: 534.0 || 378.0 || 662.0 || 511.0 || 1
label: 527.0 || 541.0 || 656.0 || 678.0 || 1
label: 527.0 || 712.0 || 654.0 || 857.0 || 1
1 label: oiltank score: tensor(1.0000) 349.36838 || 709.4552 || 481.7989 || 854.8713
2 label: oiltank score: tensor(1.0000) 528.97327 || 546.3467 || 656.2798 || 676.9037
3 label: oiltank score: tensor(1.0000) 88.82061 || 667.3125 || 220.63167 || 812.9001
4 label: oiltank score: tensor(1.0000) 358.38913 || 373.05368 || 489.21268 || 510.5563
5 label: oiltank score: tensor(1.0000) 519.9066 || 709.3708 || 658.5541 || 863.88947
6 label: oiltank score: tensor(1.0000) 115.77756 || 339.24872 || 240.00787 || 466.19022
7 label: oiltank score: tensor(1.0000) 104.77564 || 500.64545 || 230.20764 || 636.5293
8 label: oiltank score: tensor(1.0000) 357.70694 || 547.2647 || 485.0283 || 679.9714
9 label: oiltank score: tensor(0.9999) 524.7541 || 375.12167 || 668.46106 || 511.2381
打印测试结果
import os
import sys
module_path = os.path.abspath(os.path.join('..'))
if module_path not in sys.path:
sys.path.append(module_path)
import torch
import torch.nn as nn
import torch.backends.cudnn as cudnn
from torch.autograd import Variable
import numpy as np
import cv2
if torch.cuda.is_available():
torch.set_default_tensor_type('torch.cuda.FloatTensor')
from ssd import build_ssd
net = build_ssd('test', 300, 3) # initialize SSD
net.load_state_dict(torch.load('weights/ssd300_COCO_60000.pth'))#加载模型
net.eval()
image = cv2.imread('data/VOCdevkit/aircraft_27.jpg', cv2.IMREAD_COLOR) # 加载图片
from matplotlib import pyplot as plt
rgb_image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)#读取图片
# View the sampled input image before transform
x = cv2.resize(image, (300, 300)).astype(np.float32)
x -= (104.0, 117.0, 123.0)
x = x.astype(np.float32)
x = x[:, :, ::-1].copy()
plt.imshow(x)
x = torch.from_numpy(x).permute(2, 0, 1)
xx = Variable(x.unsqueeze(0)) # wrap tensor in Variable
xx = xx.cuda()
y = net(xx)
from data import VOC_CLASSES as labels
top_k=10#选择前10个结果
plt.figure(figsize=(10,10))
colors = plt.cm.hsv(np.linspace(0, 1, 21)).tolist()
plt.imshow(rgb_image) # plot the image for matplotlib
currentAxis = plt.gca()
detections = y.data
# scale each detection back up to the image
scale = torch.Tensor(rgb_image.shape[1::-1]).repeat(2)
for i in range(detections.size(1)):
j = 0
while detections[0,i,j,0] >= 0.35:
score = detections[0,i,j,0]
label_name = labels[i-1]
display_txt = '%s: %.2f'%(label_name, score)
pt = (detections[0,i,j,1:]*scale).cpu().numpy()
coords = (pt[0], pt[1]), pt[2]-pt[0]+1, pt[3]-pt[1]+1
color = colors[i]
currentAxis.add_patch(plt.Rectangle(*coords, fill=False, edgecolor=color, linewidth=2))
currentAxis.text(pt[0], pt[1], display_txt, bbox={'facecolor':color, 'alpha':0.5})
j+=1
plt.show()