def __init__(self, cfg: CfgNode):
"""
Initialize chart-based loss from configuration options
Args:
cfg (CfgNode): configuration options
"""
# fmt: off
self.heatmap_size = cfg.MODEL.ROI_DENSEPOSE_HEAD.HEATMAP_SIZE
self.w_points = cfg.MODEL.ROI_DENSEPOSE_HEAD.POINT_REGRESSION_WEIGHTS
self.w_part = cfg.MODEL.ROI_DENSEPOSE_HEAD.PART_WEIGHTS
self.use_part_focal_loss = cfg.MODEL.ROI_DENSEPOSE_HEAD.PART_FOCAL_LOSS
self.w_segm = cfg.MODEL.ROI_DENSEPOSE_HEAD.INDEX_WEIGHTS
self.w_body = cfg.MODEL.ROI_DENSEPOSE_HEAD.BODY_WEIGHTS
self.n_segm_chan = cfg.MODEL.ROI_DENSEPOSE_HEAD.NUM_COARSE_SEGM_CHANNELS
self.w_smooth = cfg.MODEL.ROI_DENSEPOSE_HEAD.SMOOTH_WEIGHTS
self.w_tv = cfg.MODEL.ROI_DENSEPOSE_HEAD.TV_WEIGHTS
# fmt: on
self.segm_trained_by_masks = cfg.MODEL.ROI_DENSEPOSE_HEAD.COARSE_SEGM_TRAINED_BY_MASKS
# <<<<<<< HEAD
self.use_mean_uv = cfg.MODEL.ROI_DENSEPOSE_HEAD.MEAN_UV_LOSS
if self.use_part_focal_loss:
gamma = cfg.MODEL.ROI_DENSEPOSE_HEAD.PART_FOCAL_GAMMA
self.focal_loss = FocalLoss(gamma=gamma,
alpha=None,
size_average=False)
self.use_teacher_student = cfg.MODEL.TEACHER_STUDENT
self.teacher_cfg = cfg.MODEL.TEACHER_CFG_FILE
self.teacher_weights = cfg.MODEL.TEACHER_WEIGHTS
self.teach_ins_wo_gt_dp = cfg.MODEL.TEACH_INS_WO_GT_DP
self.w_part_teach = cfg.MODEL.TEACH_PART_WEIGHTS
self.w_points_teach = cfg.MODEL.TEACH_POINT_REGRESSION_WEIGHTS
if self.use_teacher_student:
from densepose.engine import Trainer
from densepose.modeling.densepose_checkpoint import DensePoseCheckpointer
from densepose.config import get_cfg, add_densepose_config
self.teacher_cfg = get_cfg()
add_densepose_config(self.teacher_cfg)
self.teacher_cfg.merge_from_file(cfg.MODEL.TEACHER_CFG_FILE)
self.teacher_model = Trainer.build_model(self.teacher_cfg)
# pdb.set_trace()
DensePoseCheckpointer(self.teacher_model).load(
cfg.MODEL.TEACHER_WEIGHTS)
self.teacher_model.eval()
self.use_aux_global_s = cfg.MODEL.CONDINST.AUX_SUPERVISION_GLOBAL_S
self.use_aux_global_skeleton = cfg.MODEL.CONDINST.AUX_SUPERVISION_GLOBAL_SKELETON
self.use_aux_body_semantics = cfg.MODEL.CONDINST.AUX_SUPERVISION_BODY_SEMANTICS
self.w_aux_global_s = cfg.MODEL.CONDINST.AUX_SUPERVISION_GLOBAL_S_WEIGHTS
self.w_aux_global_skeleton = cfg.MODEL.CONDINST.AUX_SUPERVISION_GLOBAL_SKELETON_WEIGHTS
self.w_aux_body_semantics = cfg.MODEL.CONDINST.AUX_SUPERVISION_BODY_SEMANTICS
self.pred_ins_body = cfg.MODEL.CONDINST.PREDICT_INSTANCE_BODY
# def __call__(
# self, proposals_with_gt: List[Instances], densepose_predictor_outputs: Any, images=None
# =======
self.segm_loss = MaskOrSegmentationLoss(cfg)
def setup(args):
cfg = get_cfg()
add_densepose_config(cfg)
cfg.merge_from_file(args.config_file)
cfg.merge_from_list(args.opts)
cfg.freeze()
default_setup(cfg, args)
# Setup logger for "densepose" module
setup_logger(output=cfg.OUTPUT_DIR, distributed_rank=comm.get_rank(), name="densepose")
return cfg
def setup_config(cls: type, config_fpath: str, model_fpath: str,
args: argparse.Namespace, opts: List[str]):
cfg = get_cfg()
add_densepose_config(cfg)
add_hrnet_config(cfg)
cfg.merge_from_file(config_fpath)
cfg.merge_from_list(args.opts)
if opts:
cfg.merge_from_list(opts)
cfg.MODEL.WEIGHTS = model_fpath
## MLQ added
cfg.MODEL.DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
cfg.DATA_DIR = os.path.dirname(args.input)
# cfg.SMOOTH_K = args.smooth_k
cfg.freeze()
return cfg
def setup_config(cls: type, config_fpath: str, model_fpath: str,
args: argparse.Namespace, opts: List[str]):
# pdb.set_trace()
cfg = get_cfg()
add_densepose_config(cfg)
# add_hrnet_config(cfg)
# add_adet_cfg(cfg, args)
cfg.merge_from_file(config_fpath)
cfg.merge_from_list(args.opts)
if opts:
cfg.merge_from_list(opts)
cfg.MODEL.WEIGHTS = model_fpath
## MLQ added
cfg.MODEL.DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
cfg.DATA_DIR = os.path.dirname(args.input)
cfg.SMOOTH_K = args.smooth_k
# Set score_threshold for builtin models
cfg.MODEL.RETINANET.SCORE_THRESH_TEST = args.confidence_threshold
cfg.MODEL.ROI_HEADS.SCORE_THRESH_TEST = args.confidence_threshold
cfg.MODEL.FCOS.INFERENCE_TH_TEST = args.confidence_threshold
cfg.MODEL.MEInst.INFERENCE_TH_TEST = args.confidence_threshold
cfg.MODEL.PANOPTIC_FPN.COMBINE.INSTANCES_CONFIDENCE_THRESH = args.confidence_threshold
cfg.freeze()
return cfg