model.load_state_dict(torch.load(opt.model_path, map_location=torch.device('cpu')))
model.eval()
close = False
for dataset in ['COD10K-v3', 'MYTEST', 'CAMO', 'CHAMELEON', 'COD10K']:
if close:
break
save_path = opt.test_save + dataset + '/'
os.makedirs(save_path, exist_ok=True)
imgpath = './Dataset/TestDataset/{}/Imgs/'.format(dataset)
gtpath = './Dataset/TestDataset/{}/GT/'.format(dataset)
if dataset == 'COD10K-v3':
imgpath = './Dataset/TestDataset/{}/Image/'.format(dataset)
gtpath = './Dataset/TestDataset/{}/GT_Instance/'.format(dataset)
test_loader = test_dataset(imgpath, gtpath, opt.testsize)
img_count = 1
for iteration in range(test_loader.size):
start_time = time.time()
image, gt, name = test_loader.load_data()
gt = np.asarray(gt, np.float32)
gt /= (gt.max() + 1e-8)
_, cam = model(image)
cam = F.interpolate(cam, size=gt.shape, mode='bilinear', align_corners=True)
cam = cam.sigmoid().data.cpu().numpy().squeeze()
# normalize
cam = (cam - cam.min()) / (cam.max() - cam.min() + 1e-8)
parser.add_argument('--model_path',
type=str,
default='./Snapshot/FeaNet/SINet_Final.pth')
parser.add_argument('--test_save', type=str, default='./Result/FeaNet/')
opt = parser.parse_args()
model = SINet_ResNet50().cuda()
model.load_state_dict(torch.load(opt.model_path))
model.eval()
for dataset in ['COD10K', 'CAMO', 'CHAMELEON']:
save_path = opt.test_save + dataset + '/'
os.makedirs(save_path, exist_ok=True)
test_loader = test_dataset(
'./Dataset/TestDataset/{}/Image/'.format(dataset),
'./Dataset/TestDataset/{}/GT/'.format(dataset), opt.testsize)
img_count = 0
totalMae = 0
for iteration in range(test_loader.size):
image, gt, name = test_loader.load_data()
gt = np.asarray(gt, np.float32)
gt /= (gt.max() + 1e-8)
image = image.cuda()
cam, _ = model(image)
cam = F.upsample(cam,
size=gt.shape,
mode='bilinear',
default='/content/drive/MyDrive/SINet_40.pth')
parser.add_argument('--test_save',
type=str,
default='/content/drive/MyDrive/Test/')
opt = parser.parse_args()
model = SINet_ResNet50().cuda()
model.load_state_dict(torch.load(opt.model_path))
model.eval()
for dataset in ['Test10']:
save_path = opt.test_save + dataset + '/'
os.makedirs(save_path, exist_ok=True)
test_loader = test_dataset(
image_root='/content/drive/MyDrive/video10/'.format(dataset),
gt_root='/content/drive/MyDrive/Tvideo10/'.format(dataset),
testsize=opt.testsize)
img_count = 1
f = np.zeros(test_loader.size)
for iteration in range(test_loader.size):
# load data
image, gt, name = test_loader.load_data()
gt = np.asarray(gt, np.float32)
gt /= (gt.max() + 1e-8)
image = image.cuda()
# inference
_, cam = model(image)
# reshape and squeeze
cam = F.upsample(cam,
size=gt.shape,
opt = parser.parse_args()
model = SINet_ResNet50().cuda()
model.load_state_dict(torch.load(opt.model_path))
model.eval()
for dataset in ['COD10K']:
save_path = opt.test_save + dataset + '/'
os.makedirs(save_path, exist_ok=True)
# NOTES:
# if you plan to inference on your customized dataset without grouth-truth,
# you just modify the params (i.e., `image_root=your_test_img_path` and `gt_root=your_test_img_path`)
# with the same filepath. We recover the original size according to the shape of grouth-truth, and thus,
# the grouth-truth map is unnecessary actually.
test_loader = test_dataset(
image_root='./Dataset/TestDataset/{}/Image/'.format(dataset),
gt_root='./Dataset/TestDataset/{}/GT/'.format(dataset),
testsize=opt.testsize)
img_count = 1
for iteration in range(test_loader.size):
# load data
image, gt, name = test_loader.load_data()
gt = np.asarray(gt, np.float32)
gt /= (gt.max() + 1e-8)
image = image.cuda()
# inference
_, cam = model(image)
# reshape and squeeze
cam = F.upsample(cam,
size=gt.shape,
mode='bilinear',
align_corners=True)
model = SINet_ResNet50().cuda()
model.load_state_dict(torch.load(opt.model_path))
model.eval()
for dataset in ['COD10K']:
#for dataset in ['COD10K-v3/Test']:
save_path = opt.test_save + dataset + '/'
os.makedirs(save_path, exist_ok=True)
# NOTES:
# if you plan to inference on your customized dataset without grouth-truth,
# you just modify the params (i.e., `image_root=your_test_img_path` and `gt_root=your_test_img_path`)
# with the same filepath. We recover the original size according to the shape of grouth-truth, and thus,
# the grouth-truth map is unnecessary actually.
test_loader = test_dataset(image_root=image_root.format(dataset),
gt_root=gt_root.format(dataset),
testsize=opt.testsize)
img_count = 1
for iteration in range(test_loader.size):
# load data
image, gt, name = test_loader.load_data()
gt = np.asarray(gt, np.float32)
gt /= (gt.max() + 1e-8)
image = image.cuda()
# inference
_, cam = model(image)
# reshape and squeeze
cam = F.upsample(cam,
size=gt.shape,
mode='bilinear',
align_corners=True)