lr = opt.lr * (gamma ** (step))
print('change learning rate, now learning rate is :', lr)
for param_group in optimizer.param_groups:
param_group['lr'] = lr
if __name__ == '__main__':
print_config('config.py')
print('now runing on device : ', device)
if not os.path.exists(opt.save_folder):
os.mkdir(opt.save_folder)
model = SSD(opt.num_classes, opt.anchor_num)
if opt.resume:
print('loading checkpoint...')
model.load_state_dict(torch.load(opt.resume))
else:
vgg_weights = torch.load(opt.save_folder + opt.basenet)
print('Loading base network...')
model.vgg.load_state_dict(vgg_weights)
model.to(device)
model.train()
mb = MultiBoxEncoder(opt)
image_sets = [['2007', 'trainval'], ['2012', 'trainval']]
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
voc_root = opt.VOC_ROOT
annopath = os.path.join(voc_root, 'VOC2007', 'Annotations',
"%s.xml") #真实的标签坐标所在的地方
imgpath = os.path.join(voc_root, 'VOC2007', 'JPEGImages', '%s.jpg') #图片所在的地方
imgsetpath = os.path.join(voc_root, 'VOC2007', 'ImageSets', 'Main',
'{:s}.txt') #测试集的txt文件存在的地方
cachedir = os.path.join(os.getcwd(), 'annotations_cache') #暂存这些信息的地方
if __name__ == '__main__':
print('using {} to eval, use cpu may take an hour to complete !!'.format(
device))
model = SSD(opt.num_classes, opt.anchor_num)
print('loading checkpoint from {}'.format(checkpoint))
state_dict = torch.load(
checkpoint, map_location=None if torch.cuda.is_available() else 'cpu')
model.load_state_dict(state_dict)
model.to(device)
print('model loaded')
multibox_encoder = MultiBoxEncoder(opt)
image_sets = [['2007', 'test']]
test_dataset = VOCDetection(opt, image_sets=image_sets, is_train=False)
os.makedirs(output_dir, exist_ok=True)
"""
lr = opt.lr * (gamma**(step))
print('change learning rate, now learning rate is :', lr)
for param_group in optimizer.param_groups:
param_group['lr'] = lr
if __name__ == '__main__':
print_config('config.py')
print('now runing on device : ', device)
if not os.path.exists(opt.save_folder):
os.mkdir(opt.save_folder)
model = SSD(opt.num_classes, opt.anchor_num)
if opt.resume:
print('loading checkpoint...')
model.load_state_dict(torch.load(opt.resume))
else:
vgg_weights = torch.load(opt.save_folder + opt.basenet)
print('Loading base network...')
model.vgg.load_state_dict(vgg_weights)
model.to(device)
model.train()
mb = MultiBoxEncoder(opt)
image_sets = [['2007', 'trainval'], ['2012', 'trainval']]
dataset = CustomDetection(