def xsnet1x34(pretrained_model=None, **kwargs):
model = XSNet(Block, [4, 4, 8], [32 * 2, 64 * 2, 96 * 2], **kwargs)
# model = XSNet(Block, [2, 4, 4], [32, 64, 96], **kwargs)
if pretrained_model is not None:
model = load_model(model, pretrained_model)
return model
def Seg_Model(num_classes, criterion=None, pretrained_model=None, recurrence=0, **kwargs):
model = ResNet(Bottleneck,[3, 4, 23, 3], num_classes, criterion, recurrence)
if pretrained_model is not None:
model = load_model(model, pretrained_model)
return model
def Seg_Model(num_classes, criterion=None, pretrained_model=None):
model = ResNet(Bottleneck, [3, 4, 23, 3], num_classes, criterion)
if pretrained_model is not None:
model = load_model(model, pretrained_model)
return model
def iwxsnet18(pretrained_model=None, **kwargs):
model = XSNet(Block, [2, 2, 4], [32, 64, 96], **kwargs)
# model = XSNet(Block, [2, 4, 4], [32, 64, 96], **kwargs)
if pretrained_model is not None:
model = load_model(model, pretrained_model)
return model
def set_ckpt(self, path):
self.save_pth = path
try:
self.net = load_model(self.net, self.save_pth)
self.signal_message.emit(int(2))
except RuntimeError:
self.signal_message.emit(int(1))
def set_net(self, net):
self.net_num = net
print(net)
self.init_net()
self.net = load_model(self.net, self.save_pth)
self.net.eval()
if self.is_cuda:
self.net.cuda()
def run(self, model_path, model_indice, log_file, log_file_link):
"""There are four evaluation modes:
1.only eval a .pth model: -e *.pth
2.only eval a certain epoch: -e epoch
3.eval all epochs in a given section: -e start_epoch-end_epoch
4.eval all epochs from a certain started epoch: -e start_epoch-
"""
if '.pth' in model_indice:
models = [
model_indice,
]
elif "-" in model_indice:
start_epoch = int(model_indice.split("-")[0])
end_epoch = model_indice.split("-")[1]
models = os.listdir(model_path)
models.remove("epoch-last.pth")
sorted_models = [None] * len(models)
model_idx = [0] * len(models)
for idx, m in enumerate(models):
num = m.split(".")[0].split("-")[1]
model_idx[idx] = num
sorted_models[idx] = m
model_idx = np.array([int(i) for i in model_idx])
down_bound = model_idx >= start_epoch
up_bound = [True] * len(sorted_models)
if end_epoch:
end_epoch = int(end_epoch)
assert start_epoch < end_epoch
up_bound = model_idx <= end_epoch
bound = up_bound * down_bound
model_slice = np.array(sorted_models)[bound]
models = [os.path.join(model_path, model) for model in model_slice]
else:
models = [
os.path.join(model_path, 'epoch-%s.pth' % model_indice),
]
results = open(log_file, 'a')
link_file(log_file, log_file_link)
for model in models:
logger.info("Load Model: %s" % model)
self.val_func = load_model(self.network, model)
# for name, parameters in self.val_func.named_parameters():
# print(name, ':', parameters.size())
result_line = self.multi_process_evaluation()
results.write('Model: ' + model + '\n')
results.write(result_line)
results.write('\n')
results.flush()
results.close()
def xsnet18v2(pretrained_model=None, **kwargs):
model = XSNet(Block, [2, 2, 4], [32, 64, 96],
context_embedding=ContextEmbeddingV2,
**kwargs)
# model = XSNet(Block, [2, 4, 4], [32, 64, 96], **kwargs)
if pretrained_model is not None:
model = load_model(model, pretrained_model)
return model
def resnet152(pretrained_model=None, **kwargs):
model = ResNet(Bottleneck, [3, 8, 36, 3], **kwargs)
if pretrained_model is not None:
model = load_model(model, pretrained_model)
return model
# if __name__ == "__main__":
# res = resnet101()
# print(res)
def set_dataset(self, dataset):
self.dataset = dataset
if self.dataset == 0:
self.color_map = self.color_map_fish
elif self.dataset == 1:
self.color_map = self.color_map_city
self.init_net()
self.net = load_model(self.net, self.save_pth)
self.net.eval()
if self.is_cuda:
self.net.cuda()
def __init__(self,is_cuda):
super(Inference, self).__init__()
self.mode = 0
self.video_path = 'city.avi'
self.camera_num = 0
self.source = 0
self.stop = False
self.is_cuda = is_cuda
self.dataset = 0 # Fish-0 city-1
self.net_num = 0
self.num_classes = 18
self.input_path = './res/1_Img8bit.png'
self.label_path = './res/1_gtFine_labelTrainIds.png'
self.gt_path = './res/1_gtFine_color.png'
mean = (104.00698793, 116.66876762, 122.67891434)
respth = './res'
self.color_map_fish = [[128, 64, 128], [250, 170, 160], [250, 170, 30], [220, 220, 0], [153, 153, 153], [180, 165, 180],
[243, 35, 232], [220, 20, 59], [254, 0, 0], [0, 0, 142], [0, 0, 70],
[1, 60, 100], [0, 0, 230], [119, 12, 32], [70, 70, 70], [107, 142, 35], [153, 251, 152],
[70, 130, 180]]
self.color_map_fish.append([0, 0, 0])
self.color_map_fish = np.array(self.color_map_fish)
with open('./cityscapes_info.json', 'r') as fr:
labels_info = json.load(fr)
self.lb_map = {el['id']: el['trainId'] for el in labels_info}
color = {el['trainId']: el['color'] for el in labels_info}
self.color_map_city = []
for i in range(19):
self.color_map_city.append(color[i])
self.color_map_city.append([0, 0, 0])
self.color_map_city = np.array(self.color_map_city)
self.save_pth = osp.join(respth, 'FT.pth')
self.init_net()
self.net = load_model(self.net, self.save_pth)
self.net.eval()
if self.is_cuda:
self.net.cuda()
def Seg_Model(num_classes, criterion=None, pretrained_model=None):
model = ResNet(Bottleneck, [3, 4, 23, 3], num_classes, criterion)
if pretrained_model is not None:
model = load_model(model, pretrained_model)
# device = torch.device('cpu')
# saved_state_dict = torch.load(pretrained_model, map_location=device)
# new_params = model.state_dict().copy()
# for i in saved_state_dict:
# #Scale.layer5.conv2d_list.3.weight
# i_parts = i.split('.')
# # print i_parts
# # if not i_parts[1]=='layer5':
# if not i_parts[0]=='fc':
# new_params['.'.join(i_parts[0:])] = saved_state_dict[i]
# model.load_state_dict(new_params)
return model
def restore_checkpoint(self):
t_start = time.time()
if self.distributed:
tmp = torch.load(self.continue_state_object,
map_location=lambda storage, loc: storage.cuda(
self.local_rank))
else:
tmp = torch.load(self.continue_state_object)
t_ioend = time.time()
self.state.model = load_model(self.state.model, tmp['model'], True)
self.state.optimizer.load_state_dict(tmp['optimizer'])
self.state.epoch = tmp['epoch'] + 1
self.state.iteration = tmp['iteration']
del tmp
t_end = time.time()
logger.info("Load checkpoint from file {}, "
"Time usage:\n\tIO: {}, restore snapshot: {}".format(
self.continue_state_object, t_ioend - t_start,
t_end - t_ioend))
def run(self, model_path, model_indice, log_file, log_file_link):
"""Evaluate models."""
if '.pth' in model_indice:
models = [model_indice, ]
else:
models = [os.path.join(model_path,
'epoch-%s.pth' % model_indice), ]
results = open(log_file, 'a')
link_file(log_file, log_file_link)
for model in models:
logger.info("Load Model: %s" % model)
self.val_func = load_model(self.network, model)
result_line = self.multi_process_evaluation()
results.write('Model: ' + model + '\n')
results.write(result_line)
results.write('\n')
results.flush()
results.close()
def resnet34(pretrained_model=None, **kwargs):
model = ResNet(BasicBlock, [3, 4, 6, 3], **kwargs)
if pretrained_model is not None:
model = load_model(model, pretrained_model)
return model
def xception39(pretrained_model=None, **kwargs):
model = Xception(Block, [4, 8, 4], [16, 32, 64], **kwargs)
if pretrained_model is not None:
model = load_model(model, pretrained_model)
return model
def xsnet18_v1b(pretrained_model=None, **kwargs):
model = XSNet(Block, [2, 3, 3], [32, 64, 96], **kwargs)
if pretrained_model is not None:
model = load_model(model, pretrained_model)
return model
def main():
"""Create the model and start the evaluation process."""
parser = get_parser()
with Engine(custom_parser=parser) as engine:
args = parser.parse_args()
cudnn.benchmark = True
h, w = map(int, args.input_size.split(','))
if args.whole:
input_size = (1024, 2048)
else:
input_size = (h, w)
seg_model = eval('networks.' + args.model + '.Seg_Model')(
num_classes=args.num_classes, recurrence=args.recurrence)
load_model(seg_model, args.restore_from)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
seg_model.to(device)
model = engine.data_parallel(seg_model)
model.eval()
dataset = CSDataSet(args.data_dir,
args.data_list,
crop_size=(1024, 2048),
mean=IMG_MEAN,
scale=False,
mirror=False)
test_loader, test_sampler = engine.get_test_loader(dataset)
if engine.distributed:
test_sampler.set_epoch(0)
data_list = []
confusion_matrix = np.zeros((args.num_classes, args.num_classes))
palette = get_palette(256)
save_path = os.path.join(os.path.dirname(args.restore_from), 'outputs')
if not os.path.exists(save_path):
os.makedirs(save_path)
bar_format = '{desc}[{elapsed}<{remaining},{rate_fmt}]'
pbar = tqdm(range(len(test_loader)),
file=sys.stdout,
bar_format=bar_format)
dataloader = iter(test_loader)
for idx in pbar:
image, label, size, name = dataloader.next()
size = size[0].numpy()
with torch.no_grad():
output = predict_multiscale(model, image, input_size, [1.0],
args.num_classes, False, 0)
seg_pred = np.asarray(np.argmax(output, axis=3), dtype=np.uint8)
seg_gt = np.asarray(label.numpy()[:, :size[0], :size[1]],
dtype=np.int)
for i in range(image.size(0)):
output_im = PILImage.fromarray(seg_pred[i])
output_im.putpalette(palette)
output_im.save(os.path.join(save_path, name[i] + '.png'))
ignore_index = seg_gt != 255
seg_gt = seg_gt[ignore_index]
seg_pred = seg_pred[ignore_index]
# show_all(gt, output)
confusion_matrix += get_confusion_matrix(seg_gt, seg_pred,
args.num_classes)
print_str = ' Iter{}/{}'.format(idx + 1, len(test_loader))
pbar.set_description(print_str, refresh=False)
confusion_matrix = torch.from_numpy(
confusion_matrix).contiguous().cuda()
confusion_matrix = engine.all_reduce_tensor(confusion_matrix,
norm=False).cpu().numpy()
pos = confusion_matrix.sum(1)
res = confusion_matrix.sum(0)
tp = np.diag(confusion_matrix)
IU_array = (tp / np.maximum(1.0, pos + res - tp))
mean_IU = IU_array.mean()
# getConfusionMatrixPlot(confusion_matrix)
if engine.distributed and engine.local_rank == 0:
print({'meanIU': mean_IU, 'IU_array': IU_array})
model_path = os.path.dirname(args.restore_from)
with open(os.path.join(model_path, 'result.txt'), 'w') as f:
f.write(
json.dumps({
'meanIU': mean_IU,
'IU_array': IU_array.tolist()
}))
def resnet101_new(pretrained_model=None, **kwargs):
model = ResNet_new(Bottleneck, [3, 4, 23, 3], **kwargs)
if pretrained_model is not None:
model = load_model(model, pretrained_model)
return model
def resnet152(pretrained_model=None, **kwargs):
model = ResNet(Bottleneck, [3, 8, 36, 3], **kwargs)
if pretrained_model is not None:
model = load_model(model, pretrained_model)
return model
from network import MCNet
from utils.pyt_utils import load_model
from tools.benchmark.compute_speed import compute_speed
from tools.benchmark.compute_flops import compute_flops
from tools.benchmark.flops_params_count import get_model_complexity_info
if __name__ == "__main__":
network = MCNet(10, criterion=None, edge_criterion=None)
model_file = "/home/xionghaitao/workplace/segmantic_segmentation/TorchSeg/log/scut.ernet.R101/snapshot/epoch-49.pth"
model = load_model(network, model_file)
model = model.cuda()
model.eval()
device = 0
flops_count, params_count = get_model_complexity_info(model, (3, 576, 720))
def xception71(pretrained_model=None, **kwargs):
model = Xception(**kwargs)
if pretrained_model is not None:
model = load_model(model, pretrained_model)
return model
