def validation(model, data_path):
model.eval()
image_root = f'{data_path}/images/'
gt_root = f'{data_path}/masks/'
validation_loader = ValidationDataset(image_root, gt_root, 352)
b = 0.0
for i in range(validation_loader.size):
image, gt, name = validation_loader.load_data()
gt = np.asarray(gt, np.float32)
gt /= (gt.max() + 1e-8)
image = image.cuda()
res = model(image)
res = F.upsample(res,
size=gt.shape,
mode='bilinear',
align_corners=True)
res = res.sigmoid().data.cpu().numpy().squeeze()
res = (res - res.min()) / (res.max() - res.min() + 1e-8)
input_ = res
target = np.array(gt)
N = gt.shape
smooth = 1
input_flat = np.reshape(input_, (-1))
target_flat = np.reshape(target, (-1))
intersection = (input_flat * target_flat)
loss = (2 * intersection.sum() + smooth) / (input_.sum() +
target.sum() + smooth)
a = '{:4f}'.format(loss)
a = float(a)
b = b + a
return b / validation_loader.size
pth_path = config['Paths']['effnetv2pd_pth_path']
result_path = config['Paths']['effnetv2pd_result_path']
model.load_state_dict(torch.load(pth_path))
model.cuda()
model.eval()
for _data_name in [
'CVC-300', 'CVC-ClinicDB', 'Kvasir', 'CVC-ColonDB', 'ETIS-LaribPolypDB'
]:
data_path = f'{test_path}/{_data_name}'
save_path = f'{result_path}/{_data_name}/'
os.makedirs(save_path, exist_ok=True)
image_root = '{}/images/'.format(data_path)
gt_root = '{}/masks/'.format(data_path)
test_loader = ValidationDataset(image_root, gt_root, test_size)
for i 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()
if model_name == 'pranet' or model_name == 'prahardnet':
_, _, _, res = model(image)
elif model_name == 'hardmseg' or model_name == 'resnetpd' or model_name == 'effnetv2pd':
res = model(image)
elif model_name == 'hardcpd' or model_name == 'resnetcpd' or model_name == 'effnetv2cpd':
_, res = model(image)
res = F.upsample(res,
size=gt.shape,