def test_seg_model(args):
if args.model_name == "UNet":
model = UNet(n_channels=args.in_channels, n_classes=args.class_num)
elif args.model_name == "PSP":
model = pspnet.PSPNet(n_classes=19, input_size=(448, 448))
model.classification = nn.Conv2d(512, args.class_num, kernel_size=1)
else:
raise NotImplemented("Unknown model {}".format(args.model_name))
model_path = os.path.join(args.model_dir, args.best_model)
model = nn.DataParallel(model)
model.load_state_dict(torch.load(model_path))
model.cuda()
model.eval()
print('--------Start testing--------')
since = time.time()
dloader = gen_dloader(os.path.join(args.data_dir, "val"),
args.batch_size,
mode="val")
metrics = defaultdict(float)
ttl_samples = 0
# preds_dir = os.path.join(args.data_dir, "val/preds", args.model_name)
# filesystem.overwrite_dir(preds_dir)
for batch_ind, (imgs, masks) in enumerate(dloader):
if batch_ind != 0 and batch_ind % 100 == 0:
print("Processing {}/{}".format(batch_ind, len(dloader)))
inputs = Variable(imgs.cuda())
masks = Variable(masks.cuda())
with torch.no_grad():
outputs = model(inputs)
loss = calc_loss(outputs, masks, metrics)
# result_img = gen_patch_pred(inputs, masks, outputs)
# result_path = os.path.join(preds_dir, str(uuid.uuid1())[:8] + ".png")
# io.imsave(result_path, result_img)
ttl_samples += inputs.size(0)
avg_dice = metrics['dice'] / ttl_samples
time_elapsed = time.time() - since
print('Testing takes {:.0f}m {:.2f}s'.format(time_elapsed // 60,
time_elapsed % 60))
print("----Dice coefficient is: {:.3f}".format(avg_dice))
def load_seg_model(args):
if args.seg_model_name == "UNet":
seg_model = UNet(n_channels=args.in_channels,
n_classes=args.seg_class_num)
elif args.seg_model_name == "PSP":
seg_model = pspnet.PSPNet(n_classes=19,
input_size=(args.patch_len, args.patch_len))
seg_model.classification = nn.Conv2d(512,
args.seg_class_num,
kernel_size=1)
else:
raise NotImplemented("Unknown model {}".format(args.seg_model_name))
seg_model_path = os.path.join(args.model_dir, "SegBestModel",
args.best_seg_model)
seg_model = nn.DataParallel(seg_model)
seg_model.load_state_dict(torch.load(seg_model_path))
seg_model.cuda()
seg_model.eval()
return seg_model
def train_seg_model(args):
# model
model = None
if args.model_name == "UNet":
model = UNet(n_channels=args.in_channels, n_classes=args.class_num)
elif args.model_name == "PSP":
model = pspnet.PSPNet(n_classes=19, input_size=(512, 512))
model.load_pretrained_model(
model_path="./segnet/pspnet/pspnet101_cityscapes.caffemodel")
model.classification = nn.Conv2d(512, args.class_num, kernel_size=1)
else:
raise AssertionError("Unknow modle: {}".format(args.model_name))
model = nn.DataParallel(model)
model.cuda()
# optimizer
optimizer = None
if args.optim_name == "Adam":
optimizer = optim.Adam(filter(lambda p: p.requires_grad,
model.parameters()),
lr=1.0e-3)
elif args.optim_name == "SGD":
optimizer = optim.SGD(filter(lambda p: p.requires_grad,
model.parameters()),
lr=args.init_lr,
momentum=0.9,
weight_decay=0.0005)
else:
raise AssertionError("Unknow optimizer: {}".format(args.optim_name))
scheduler = lr_scheduler.LambdaLR(optimizer,
lr_lambda=LambdaLR(args.maxepoch, 0,
0).step)
# dataloader
train_data_dir = os.path.join(args.data_dir, args.tumor_type, "train")
train_dloader = gen_dloader(train_data_dir,
args.batch_size,
mode="train",
normalize=args.normalize,
tumor_type=args.tumor_type)
test_data_dir = os.path.join(args.data_dir, args.tumor_type, "val")
val_dloader = gen_dloader(test_data_dir,
args.batch_size,
mode="val",
normalize=args.normalize,
tumor_type=args.tumor_type)
# training
save_model_dir = os.path.join(args.model_dir, args.tumor_type,
args.session)
if not os.path.exists(save_model_dir):
os.makedirs(save_model_dir)
best_dice = 0.0
for epoch in np.arange(0, args.maxepoch):
print('Epoch {}/{}'.format(epoch + 1, args.maxepoch))
print('-' * 10)
since = time.time()
for phase in ['train', 'val']:
if phase == 'train':
dloader = train_dloader
scheduler.step()
for param_group in optimizer.param_groups:
print("Current LR: {:.8f}".format(param_group['lr']))
model.train() # Set model to training mode
else:
dloader = val_dloader
model.eval() # Set model to evaluate mode
metrics = defaultdict(float)
epoch_samples = 0
for batch_ind, (imgs, masks) in enumerate(dloader):
inputs = Variable(imgs.cuda())
masks = Variable(masks.cuda())
optimizer.zero_grad()
with torch.set_grad_enabled(phase == 'train'):
outputs = model(inputs)
loss = calc_loss(outputs,
masks,
metrics,
bce_weight=args.bce_weight)
if phase == 'train':
loss.backward()
optimizer.step()
# statistics
epoch_samples += inputs.size(0)
print_metrics(metrics, epoch_samples, phase)
epoch_dice = metrics['dice'] / epoch_samples
# deep copy the model
if phase == 'val' and (epoch_dice > best_dice
or epoch > args.maxepoch - 5):
best_dice = epoch_dice
best_model = copy.deepcopy(model.state_dict())
best_model_name = "-".join([
args.model_name,
"{:03d}-{:.3f}.pth".format(epoch, best_dice)
])
torch.save(best_model,
os.path.join(save_model_dir, best_model_name))
time_elapsed = time.time() - since
print('Epoch {:2d} takes {:.0f}m {:.0f}s'.format(
epoch, time_elapsed // 60, time_elapsed % 60))
print(
"================================================================================"
)
print("Training finished...")
def test_slide_seg(args):
model = None
if args.model_name == "UNet":
model = UNet(n_channels=args.in_channels, n_classes=args.class_num)
elif args.model_name == "PSP":
model = pspnet.PSPNet(n_classes=19, input_size=(512, 512))
model.classification = nn.Conv2d(512, args.class_num, kernel_size=1)
else:
raise AssertionError("Unknow modle: {}".format(args.model_name))
model_path = os.path.join(args.model_dir, args.tumor_type, args.split, args.best_model)
model = nn.DataParallel(model)
model.load_state_dict(torch.load(model_path))
model.cuda()
model.eval()
since = time.time()
result_dir = os.path.join(args.result_dir, args.tumor_type)
filesystem.overwrite_dir(result_dir)
slide_names = get_slide_filenames(args.slides_dir)
if args.save_org and args.tumor_type == "viable":
org_result_dir = os.path.join(result_dir, "Level0")
filesystem.overwrite_dir(org_result_dir)
for num, cur_slide in enumerate(slide_names):
print("--{:02d}/{:02d} Slide:{}".format(num+1, len(slide_names), cur_slide))
metrics = defaultdict(float)
# load level-2 slide
slide_path = os.path.join(args.slides_dir, cur_slide+".svs")
if not os.path.exists(slide_path):
slide_path = os.path.join(args.slides_dir, cur_slide+".SVS")
wsi_head = pyramid.load_wsi_head(slide_path)
p_level = args.slide_level
pred_h, pred_w = (wsi_head.level_dimensions[p_level][1], wsi_head.level_dimensions[p_level][0])
slide_img = wsi_head.read_region((0, 0), p_level, wsi_head.level_dimensions[p_level])
slide_img = np.asarray(slide_img)[:,:,:3]
coors_arr = wsi_stride_splitting(pred_h, pred_w, patch_len=args.patch_len, stride_len=args.stride_len)
patch_arr, wmap = gen_patch_wmap(slide_img, coors_arr, plen=args.patch_len)
patch_dset = PatchDataset(patch_arr, mask_arr=None, normalize=args.normalize, tumor_type=args.tumor_type)
patch_loader = DataLoader(patch_dset, batch_size=args.batch_size, shuffle=False, num_workers=4, drop_last=False)
ttl_samples = 0
pred_map = np.zeros_like(wmap).astype(np.float32)
for ind, patches in enumerate(patch_loader):
inputs = Variable(patches.cuda())
with torch.no_grad():
outputs = model(inputs)
preds = F.sigmoid(outputs)
preds = torch.squeeze(preds, dim=1).data.cpu().numpy()
if (ind+1)*args.batch_size <= len(coors_arr):
patch_coors = coors_arr[ind*args.batch_size:(ind+1)*args.batch_size]
else:
patch_coors = coors_arr[ind*args.batch_size:]
for ind, coor in enumerate(patch_coors):
ph, pw = coor[0], coor[1]
pred_map[ph:ph+args.patch_len, pw:pw+args.patch_len] += preds[ind]
ttl_samples += inputs.size(0)
prob_pred = np.divide(pred_map, wmap)
slide_pred = (prob_pred > 0.5).astype(np.uint8)
pred_save_path = os.path.join(result_dir, cur_slide + "_" + args.tumor_type + ".tif")
io.imsave(pred_save_path, slide_pred*255)
if args.save_org and args.tumor_type == "viable":
org_w, org_h = wsi_head.level_dimensions[0]
org_pred = transform.resize(prob_pred, (org_h, org_w))
org_pred = (org_pred > 0.5).astype(np.uint8)
org_save_path = os.path.join(org_result_dir, cur_slide[-3:] + ".tif")
imsave(org_save_path, org_pred, compress=9)
time_elapsed = time.time() - since
print('Testing takes {:.0f}m {:.2f}s'.format(time_elapsed // 60, time_elapsed % 60))
def test_slide_seg(args):
if args.model_name == "UNet":
model = UNet(n_channels=args.in_channels, n_classes=args.class_num)
elif args.model_name == "PSP":
model = pspnet.PSPNet(n_classes=19,
input_size=(args.patch_len, args.patch_len))
model.classification = nn.Conv2d(512, args.class_num, kernel_size=1)
else:
raise NotImplemented("Unknown model {}".format(args.model_name))
model_path = os.path.join(args.model_dir, args.best_model)
model = nn.DataParallel(model)
model.load_state_dict(torch.load(model_path))
model.cuda()
model.eval()
since = time.time()
pydaily.filesystem.overwrite_dir(args.result_dir)
slide_names = [ele for ele in os.listdir(args.slides_dir) if "jpg" in ele]
ttl_pred_dice = 0.0
for num, cur_slide in enumerate(slide_names):
print("--{:2d}/{:2d} Slide:{}".format(num + 1, len(slide_names),
cur_slide))
start_time = timer()
# load slide image and mask
slide_path = os.path.join(args.slides_dir, cur_slide)
slide_img = io.imread(slide_path)
# split and predict
coors_arr = wsi_stride_splitting(slide_img.shape[0],
slide_img.shape[1],
patch_len=args.patch_len,
stride_len=args.stride_len)
wmap = np.zeros((slide_img.shape[0], slide_img.shape[1]),
dtype=np.int32)
pred_map = np.zeros_like(wmap).astype(np.float32)
patch_list, coor_list = [], []
for ic, coor in enumerate(coors_arr):
ph, pw = coor[0], coor[1]
patch_list.append(
slide_img[ph:ph + args.patch_len, pw:pw + args.patch_len] /
255.0)
coor_list.append([ph, pw])
wmap[ph:ph + args.patch_len, pw:pw + args.patch_len] += 1
if len(patch_list) == args.batch_size or ic + 1 == len(coors_arr):
patch_arr = np.asarray(patch_list).astype(np.float32)
patch_dset = PatchDataset(patch_arr)
patch_loader = DataLoader(patch_dset,
batch_size=args.batch_size,
shuffle=False,
num_workers=4,
drop_last=False)
with torch.no_grad():
pred_list = []
for patches in patch_loader:
inputs = Variable(patches.cuda())
outputs = model(inputs)
preds = F.sigmoid(outputs)
preds = torch.squeeze(preds, dim=1).data.cpu().numpy()
pred_list.append(preds)
batch_preds = np.concatenate(pred_list, axis=0)
for ind, coor in enumerate(coor_list):
ph, pw = coor[0], coor[1]
pred_map[ph:ph + args.patch_len,
pw:pw + args.patch_len] += batch_preds[ind]
patch_list, coor_list = [], []
prob_pred = np.divide(pred_map, wmap)
slide_pred = morphology.remove_small_objects(prob_pred > 0.5,
min_size=20480).astype(
np.uint8)
pred_save_path = os.path.join(args.result_dir,
os.path.splitext(cur_slide)[0] + ".png")
io.imsave(pred_save_path, slide_pred * 255)
end_time = timer()
print("Takes {}".format(
pydaily.tic.time_to_str(end_time - start_time, 'sec')))
time_elapsed = time.time() - since
print("stride-len: {} with batch-size: {}".format(args.stride_len,
args.batch_size))
print("Testing takes {:.0f}m {:.2f}s".format(time_elapsed // 60,
time_elapsed % 60))
# forward
# track history if only in train
with torch.no_grad():
outputs = model(inputs)
loss = calc_loss(outputs, labels, metrics)
# statistics
epoch_samples += inputs.size(0)
print_metrics(metrics, epoch_samples, "test")
if __name__ == '__main__':
args = set_args()
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu
# create model
model = None
if args.network == "UNet":
elif args.network == "PSP":
model = pspnet.PSPNet(n_classes=19, input_size=(160, 160))
model.load_pretrained_model(model_path="./segnet/pspnet/pspnet101_cityscapes.caffemodel")
model.classification = nn.Conv2d(512, args.class_num, kernel_size=1)
else:
raise Exception("Unknow network: {}".format(args.network))
print("Net: {} session: {} model name: {}".format(args.network, args.session, args.model_name))
model_path = os.path.join(args.model_dir, args.simu_type+args.network, args.session, args.model_name)
model.load_state_dict(torch.load(model_path))
model.cuda()
# train model
test_seg_model(model, args)
def test_slide_seg(args):
model = pspnet.PSPNet(n_classes=19,
input_size=(args.patch_len, args.patch_len))
model.classification = nn.Conv2d(512, args.class_num, kernel_size=1)
model_path = os.path.join(args.model_dir, args.best_model)
model = nn.DataParallel(model)
model.load_state_dict(torch.load(model_path))
model.cuda()
model.eval()
since = time.time()
# filesystem.overwrite_dir(args.result_dir)
slide_names = [ele for ele in os.listdir(args.slides_dir) if "jpg" in ele]
ttl_pred_dice = 0.0
for num, cur_slide in enumerate(slide_names):
metrics = defaultdict(float)
# load slide image and mask
slide_path = os.path.join(args.slides_dir, cur_slide)
slide_img = io.imread(slide_path) / 255.0
mask_path = os.path.join(args.slides_dir,
os.path.splitext(cur_slide)[0] + ".png")
mask_img = io.imread(mask_path) / 255.0
# split and predict
coors_arr = wsi_stride_splitting(slide_img.shape[0],
slide_img.shape[1],
patch_len=args.patch_len,
stride_len=args.stride_len)
wmap = np.zeros((slide_img.shape[0], slide_img.shape[1]),
dtype=np.int32)
pred_map = np.zeros_like(wmap).astype(np.float32)
patch_list, coor_list = [], []
for ic, coor in enumerate(coors_arr):
ph, pw = coor[0], coor[1]
patch_list.append(slide_img[ph:ph + args.patch_len,
pw:pw + args.patch_len])
coor_list.append([ph, pw])
wmap[ph:ph + args.patch_len, pw:pw + args.patch_len] += 1
if len(patch_list) == args.batch_size or ic + 1 == len(coors_arr):
patch_arr = np.asarray(patch_list).astype(np.float32)
patch_dset = PatchDataset(patch_arr)
patch_loader = DataLoader(patch_dset,
batch_size=args.batch_size,
shuffle=False,
num_workers=4,
drop_last=False)
with torch.no_grad():
pred_list = []
for patches in patch_loader:
inputs = Variable(patches.cuda())
outputs = model(inputs)
preds = F.sigmoid(outputs)
preds = torch.squeeze(preds, dim=1).data.cpu().numpy()
pred_list.append(preds)
batch_preds = np.concatenate(pred_list, axis=0)
for ind, coor in enumerate(coor_list):
ph, pw = coor[0], coor[1]
pred_map[ph:ph + args.patch_len,
pw:pw + args.patch_len] += batch_preds[ind]
patch_list, coor_list = [], []
prob_pred = np.divide(pred_map, wmap)
slide_pred = morphology.remove_small_objects(prob_pred > 0.5,
min_size=20480).astype(
np.uint8)
# pred_save_path = os.path.join(args.result_dir, os.path.splitext(cur_slide)[0]+".png")
# io.imsave(pred_save_path, slide_pred*255)
intersection = np.multiply(mask_img, slide_pred)
pred_dice = np.sum(intersection) / (np.sum(mask_img) +
np.sum(slide_pred) -
np.sum(intersection) + 1.0e-8)
ttl_pred_dice += pred_dice
print("--{:2d}/{:2d} Slide:{} JI:{:.3f}".format(
num + 1, len(slide_names), cur_slide, pred_dice))
time_elapsed = time.time() - since
print('Testing takes {:.0f}m {:.2f}s'.format(time_elapsed // 60,
time_elapsed % 60))
print('Slide-level average Dice coefficient is {:.3f}'.format(
ttl_pred_dice / len(slide_names)))