def main():
parser = argparse.ArgumentParser(
description="PyTorch DeeplabV3Plus Training")
parser.add_argument('config_file', help='config file path')
parser.add_argument(
"opts",
help="Modify config options using the command-line",
default=None,
nargs=argparse.REMAINDER,
)
args = parser.parse_args()
cfg = get_cfg_defaults()
cfg.merge_from_file(args.config_file)
cfg.merge_from_list(args.opts)
print(cfg)
torch.manual_seed(cfg.SYSTEM.SEED)
trainer = Trainer(cfg)
for epoch in range(cfg.TRAIN.START_EPOCH, cfg.TRAIN.EPOCHS):
trainer.training(epoch)
trainer.writer.close()
if k in state_dict:
model_dict[k] = v
state_dict.update(model_dict)
self.load_state_dict(state_dict)
def _initialize_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
# n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
# m.weight.data.normal_(0, math.sqrt(2. / n))
torch.nn.init.kaiming_normal_(m.weight)
elif isinstance(m, SynchronizedBatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
if __name__ == "__main__":
from deeplab3.config.defaults import get_cfg_defaults
cfg = get_cfg_defaults()
cfg.merge_from_list(
['MODEL.BACKBONE_ZOO', True, 'MODEL.INPUT_CHANNELS', 3])
input = torch.rand(1, 3, 512, 512)
model = MobileNetV2(cfg, BatchNorm=nn.BatchNorm2d)
output, low_level_feat = model(input)
print(output.size())
print(low_level_feat.size())
_C.MODEL.RESNETS.STEM_OUT_CHANNELS = 64 # Apply Deformable Convolution in stages # Specify if apply deform_conv on Res2, Res3, Res4, Res5 _C.MODEL.RESNETS.DEFORM_ON_PER_STAGE = [False, False, False, False] # Use True to use modulated deform_conv (DeformableV2, https://arxiv.org/abs/1811.11168); # Use False for DeformableV1. _C.MODEL.RESNETS.DEFORM_MODULATED = False # Number of groups in deformable conv. _C.MODEL.RESNETS.DEFORM_NUM_GROUPS = 1 # ---------------------------------------------------------------------------- # # Deeplab Backbone options # These options apply to networks imported from saved pytorch-deeplab-xception models # ---------------------------------------------------------------------------- # _C.MODEL.DEEPLAB = CN(get_cfg_defaults()) # ---------------------------------------------------------------------------- # # Solver # ---------------------------------------------------------------------------- # _C.SOLVER = CN() # Which optimizer to use [adamw, sgd] _C.SOLVER.OPTIMIZER = "adamw" # See detectron2/solver/build.py for LR scheduler options _C.SOLVER.LR_SCHEDULER_NAME = "WarmupMultiStepLR" _C.SOLVER.MAX_ITER = 40000 _C.SOLVER.BASE_LR = 0.001
import scipy.stats
import numpy as np
from tqdm import tqdm
from PIL import Image
from deeplab3.config.defaults import get_cfg_defaults
from dataloaders.utils import sample_distribution
from dataloaders.datasets.cityscapes import CityscapesSegmentation
from dataloaders.datasets.coco import COCOSegmentation
from dataloaders.datasets.sunrgbd import RGBDSegmentation
from dataloaders.SampleLoader import SampleLoader
city_rgbd = get_cfg_defaults()
city_rgbd.merge_from_file('configs/cityscapes_rgbd.yaml')
city_rgbd.merge_from_list([
'DATASET.ROOT', 'datasets/cityscapes/', 'DATASET.CITYSCAPES.DEPTH_DIR',
'completed_depth'
])
sunrgbd_rgbd = get_cfg_defaults()
sunrgbd_rgbd.merge_from_file('configs/sunrgbd.yaml')
sunrgbd_rgbd.merge_from_list(['DATASET.ROOT', 'datasets/SUNRGBD/'])
sunrgbd_rgbd_dist_train = sample_distribution(RGBDSegmentation(sunrgbd_rgbd,
split='train'),
n=100)
sunrgbd_rgbd_dist_test = sample_distribution(RGBDSegmentation(sunrgbd_rgbd,
split='test'),