def get_pred(batch_data):
# process data
batch_data = batch_data[0]
segSize = (batch_data['img_ori'].shape[0],
batch_data['img_ori'].shape[1])
img_resized_list = batch_data['img_data']
with torch.no_grad():
scores = torch.zeros(1, cfg.DATASET.num_class, segSize[0], segSize[1])
scores = async_copy_to(scores, gpu)
for img in img_resized_list:
feed_dict = batch_data.copy()
feed_dict['img_data'] = img
del feed_dict['img_ori']
del feed_dict['info']
feed_dict = async_copy_to(feed_dict, gpu)
# forward pass
pred_tmp = segmentation_module(feed_dict, segSize=segSize)
scores = scores + pred_tmp / len(cfg.DATASET.imgSizes)
_, pred = torch.max(scores, dim=1)
writer.write(pred)
pred = as_numpy(pred.squeeze(0).cpu())
pbar.update(1)
return pred
def test(segmentation_module, loader, gpu):
segmentation_module.eval()
pbar = tqdm(total=len(loader))
for batch_data in loader:
# process data
batch_data = batch_data[0]
segSize = (batch_data['img_ori'].shape[0],
batch_data['img_ori'].shape[1])
img_resized_list = batch_data['img_data']
with torch.no_grad():
scores = torch.zeros(1, cfg.DATASET.num_class, segSize[0],
segSize[1])
scores = async_copy_to(scores, gpu)
for img in img_resized_list:
feed_dict = batch_data.copy()
feed_dict['img_data'] = img
del feed_dict['img_ori']
del feed_dict['info']
feed_dict = async_copy_to(feed_dict, gpu)
# forward pass
pred_tmp = segmentation_module(feed_dict, segSize=segSize)
scores = scores + pred_tmp / len(cfg.DATASET.imgSizes)
_, pred = torch.max(scores, dim=1)
pred = as_numpy(pred.squeeze(0).cpu())
# visualization
visualize_result((batch_data['img_ori'], batch_data['info']), pred,
cfg)
pbar.update(1)
def huawei_seg(imgs, segmentation_module):
#
cfg.merge_from_file("/home/mind/model/config/ade20k-hrnetv2-huawei.yaml")
imgs = [imgs]
cfg.list_test = [{'fpath_img': x} for x in imgs]
# Dataset and Loader
dataset_test = InferDataset(cfg.list_test, cfg.DATASET)
loader_test = torch.utils.data.DataLoader(
dataset_test,
batch_size=1,
shuffle=False,
collate_fn=user_scattered_collate,
num_workers=5,
drop_last=True)
segmentation_module.cuda()
loader = loader_test
# Main loop
segmentation_module.eval()
pbar = tqdm(total=len(loader))
for batch_data in loader:
# process data
batch_data = batch_data[0]
segSize = (batch_data['img_ori'].shape[0],
batch_data['img_ori'].shape[1])
img_resized_list = batch_data['img_data']
with torch.no_grad():
scores = torch.zeros(1, cfg.DATASET.num_class, segSize[0],
segSize[1])
scores = async_copy_to(scores, 0)
for img in img_resized_list:
feed_dict = batch_data.copy()
feed_dict['img_data'] = img
del feed_dict['img_ori']
del feed_dict['info']
feed_dict = async_copy_to(feed_dict, 0)
# forward pass
pred_tmp = segmentation_module(feed_dict, segSize=segSize)
# print(pred_tmp.shape)#torch.Size([1, 2, 1024, 1024])
scores = scores + pred_tmp / len(cfg.DATASET.imgSizes)
_, pred = torch.max(scores, dim=1)
pred = as_numpy(pred.squeeze(0).cpu())
#
# visualize_result(
# (batch_data['img_ori'], batch_data['info']),
# pred,
# cfg
# )
pbar.update(1)
#
return pred
def test(segmentation_module, loader, device, names):
segmentation_module.eval()
pbar = tqdm(total=len(loader))
re = []
for batch_data in loader:
# process data
batch_data = batch_data[0]
segSize = (batch_data['img_ori'].shape[0],
batch_data['img_ori'].shape[1])
img_resized_list = batch_data['img_data']
with torch.no_grad():
scores = torch.zeros(1, cfg_ss.DATASET.num_class, segSize[0],
segSize[1])
if torch.cuda.is_available():
scores = async_copy_to(scores, device)
for img in img_resized_list:
feed_dict = batch_data.copy()
feed_dict['img_data'] = img
del feed_dict['img_ori']
del feed_dict['info']
if torch.cuda.is_available():
feed_dict = async_copy_to(feed_dict, device)
# forward pass
pred_tmp = segmentation_module(feed_dict, segSize=segSize)
scores = scores + pred_tmp / len(cfg_ss.DATASET.imgSizes)
_, pred = torch.max(scores, dim=1)
pred = as_numpy(pred.squeeze(0).cpu())
pred = np.int32(pred)
pixs = pred.size
uniques, counts = np.unique(pred, return_counts=True)
pred_ratios = {}
for idx in np.argsort(counts)[::-1]:
name = names[uniques[idx] + 1]
ratio = counts[idx] / pixs * 100
if ratio > 0.1:
pred_ratios[name] = ratio
re.append({
"original_img": batch_data['img_ori'],
"img_inf": batch_data['info'],
"pred_result": pred,
"pred_ratio": pred_ratios
})
pbar.update(1)
return re
def run_inference_for_single_image(self, image):
preproc_data = preprocess_image(image)
batch_data = preproc_data
seg_size = (batch_data['img_ori'].shape[0],
batch_data['img_ori'].shape[1])
img_resized_list = batch_data['img_data']
if self.gpu is not None:
with torch.no_grad():
# scores = torch.zeros(1, cfg.DATASET.num_class, seg_size[0], seg_size[1])
scores = torch.cuda.FloatTensor(1, cfg.DATASET.num_class,
seg_size[0],
seg_size[1]).fill_(0)
scores = async_copy_to(scores, self.gpu)
for img in img_resized_list:
feed_dict = {'img_data': img}
feed_dict = async_copy_to(feed_dict, self.gpu)
# forward pass
pred_tmp = self.model(feed_dict, segSize=seg_size)
scores = scores + pred_tmp / len(cfg.DATASET.imgSizes)
_, pred = torch.max(scores, dim=1)
pred = pred.squeeze(0).cpu().numpy()
else:
scores = torch.zeros(1, cfg.DATASET.num_class, seg_size[0],
seg_size[1])
for img in img_resized_list:
feed_dict = {'img_data': img}
# forward pass
pred_tmp = self.model(feed_dict, segSize=seg_size)
scores = scores + pred_tmp / len(cfg.DATASET.imgSizes)
_, pred = torch.max(scores, dim=1)
pred = pred.squeeze(0).numpy()
return pred
def evaluate(segmentation_module, loader, cfg, gpu_id, result_queue):
segmentation_module.eval()
for index, batch_data in enumerate(loader):
# process data
batch_data = batch_data[0]
seg_label = as_numpy(batch_data['seg_label'][0])
img_resized_list = batch_data['img_data']
with torch.no_grad():
segSize = (seg_label.shape[0], seg_label.shape[1])
scores = torch.zeros(1, cfg.DATASET.num_class, segSize[0],
segSize[1])
scores = async_copy_to(scores, gpu_id)
for img in img_resized_list:
feed_dict = batch_data.copy()
feed_dict['img_data'] = img
del feed_dict['img_ori']
del feed_dict['info']
feed_dict = async_copy_to(feed_dict, gpu_id)
# forward pass
scores_tmp = segmentation_module(feed_dict, segSize=segSize)
scores = scores + scores_tmp / len(cfg.DATASET.imgSizes)
_, pred = torch.max(scores, dim=1)
pred = as_numpy(pred.squeeze(0).cpu())
# calculate accuracy and SEND THEM TO MASTER
acc, pix = accuracy(pred, seg_label)
intersection, union = intersectionAndUnion(pred, seg_label,
cfg.DATASET.num_class)
result_queue.put_nowait((acc, pix, intersection, union))
# visualization
if cfg.VAL.visualize:
visualize_result(
(batch_data['img_ori'], seg_label, batch_data['info']), pred,
os.path.join(cfg.DIR, 'result'))
def huawei_seg(imgs):
parser = argparse.ArgumentParser(
description="PyTorch Semantic Segmentation Testing")
parser.add_argument(
"--cfg",
default="config/ade20k-hrnetv2-huawei.yaml",
metavar="FILE",
help="path to config file",
type=str,
)
parser.add_argument("--gpu",
default=0,
type=int,
help="gpu id for evaluation")
args = parser.parse_args()
cfg.merge_from_file(args.cfg)
cfg.MODEL.arch_encoder = cfg.MODEL.arch_encoder.lower()
cfg.MODEL.arch_decoder = cfg.MODEL.arch_decoder.lower()
# absolute paths of model weights
cfg.MODEL.weights_encoder = os.path.join(cfg.DIR,
'encoder_' + cfg.TEST.checkpoint)
cfg.MODEL.weights_decoder = os.path.join(cfg.DIR,
'decoder_' + cfg.TEST.checkpoint)
#
imgs = [imgs]
cfg.list_test = [{'fpath_img': x} for x in imgs]
torch.cuda.set_device(args.gpu)
# Network Builders
net_encoder = ModelBuilder.build_encoder(arch=cfg.MODEL.arch_encoder,
fc_dim=cfg.MODEL.fc_dim,
weights=cfg.MODEL.weights_encoder)
net_decoder = ModelBuilder.build_decoder(arch=cfg.MODEL.arch_decoder,
fc_dim=cfg.MODEL.fc_dim,
num_class=cfg.DATASET.num_class,
weights=cfg.MODEL.weights_decoder,
use_softmax=True)
crit = nn.NLLLoss(ignore_index=-1)
segmentation_module = SegmentationModule(net_encoder, net_decoder, crit)
# Dataset and Loader
dataset_test = InferDataset(cfg.list_test, cfg.DATASET)
loader_test = torch.utils.data.DataLoader(
dataset_test,
batch_size=1,
shuffle=False,
collate_fn=user_scattered_collate,
num_workers=5,
drop_last=True)
segmentation_module.cuda()
loader = loader_test
# Main loop
segmentation_module.eval()
pbar = tqdm(total=len(loader))
for batch_data in loader:
# process data
batch_data = batch_data[0]
segSize = (batch_data['img_ori'].shape[0],
batch_data['img_ori'].shape[1])
img_resized_list = batch_data['img_data']
with torch.no_grad():
scores = torch.zeros(1, cfg.DATASET.num_class, segSize[0],
segSize[1])
scores = async_copy_to(scores, args.gpu)
for img in img_resized_list:
feed_dict = batch_data.copy()
feed_dict['img_data'] = img
del feed_dict['img_ori']
del feed_dict['info']
feed_dict = async_copy_to(feed_dict, args.gpu)
# forward pass
pred_tmp = segmentation_module(feed_dict, segSize=segSize)
# print(pred_tmp.shape)#torch.Size([1, 2, 1024, 1024])
scores = scores + pred_tmp / len(cfg.DATASET.imgSizes)
_, pred = torch.max(scores, dim=1)
pred = as_numpy(pred.squeeze(0).cpu())
# visualization
visualize_result((batch_data['img_ori'], batch_data['info']), pred,
cfg)
pbar.update(1)
#
return pred
def evaluate(segmentation_module, loader, cfg, gpu):
acc_meter = AverageMeter()
intersection_meter = AverageMeter()
union_meter = AverageMeter()
time_meter = AverageMeter()
segmentation_module.eval()
pbar = tqdm(total=len(loader))
for batch_data in loader:
# process data
batch_data = batch_data[0]
seg_label = as_numpy(batch_data['seg_label'][0])
img_resized_list = batch_data['img_data']
torch.cuda.synchronize()
tic = time.perf_counter()
with torch.no_grad():
segSize = (seg_label.shape[0], seg_label.shape[1])
scores = torch.zeros(1, cfg.DATASET.num_class, segSize[0], segSize[1])
scores = async_copy_to(scores, gpu)
for img in img_resized_list:
feed_dict = batch_data.copy()
feed_dict['img_data'] = img
del feed_dict['img_ori']
del feed_dict['info']
feed_dict = async_copy_to(feed_dict, gpu)
# forward pass
scores_tmp = segmentation_module(feed_dict, segSize=segSize)
scores = scores + scores_tmp / len(cfg.DATASET.imgSizes)
_, pred = torch.max(scores, dim=1)
pred = as_numpy(pred.squeeze(0).cpu())
torch.cuda.synchronize()
time_meter.update(time.perf_counter() - tic)
# calculate accuracy
acc, pix = accuracy(pred, seg_label)
intersection, union = intersectionAndUnion(pred, seg_label, cfg.DATASET.num_class)
acc_meter.update(acc, pix)
intersection_meter.update(intersection)
union_meter.update(union)
# visualization
if cfg.VAL.visualize:
visualize_result(
(batch_data['img_ori'], seg_label, batch_data['info']),
pred,
os.path.join(cfg.DIR, 'result')
)
pbar.update(1)
# summary
iou = intersection_meter.sum / (union_meter.sum + 1e-10)
for i, _iou in enumerate(iou):
print('class [{}], IoU: {:.4f}'.format(i, _iou))
print('[Eval Summary]:')
print('Mean IoU: {:.4f}, Accuracy: {:.2f}%, Inference Time: {:.4f}s'
.format(iou.mean(), acc_meter.average()*100, time_meter.average()))
