def __init__(self, cfg):
self.cls_loss_func = SigmoidFocalLoss(cfg.MODEL.FCOS.LOSS_GAMMA,
cfg.MODEL.FCOS.LOSS_ALPHA)
# we make use of IOU Loss for bounding boxes regression,
# but we found that L1 in log scale can yield a similar performance
self.box_reg_loss_func = IOULoss()
self.centerness_loss_func = nn.BCEWithLogitsLoss()
def __init__(self, cfg, box_coder):
self.cfg = cfg
self.cls_loss_func = SigmoidFocalLoss(cfg.MODEL.ATSS.LOSS_GAMMA,
cfg.MODEL.ATSS.LOSS_ALPHA)
self.centerness_loss_func = nn.BCEWithLogitsLoss(reduction="sum")
self.matcher = Matcher(cfg.MODEL.ATSS.FG_IOU_THRESHOLD,
cfg.MODEL.ATSS.BG_IOU_THRESHOLD, True)
self.box_coder = box_coder
def __init__(self, cfg):
self.cls_loss_func = SigmoidFocalLoss(cfg.MODEL.FCOS.LOSS_GAMMA,
cfg.MODEL.FCOS.LOSS_ALPHA)
self.fpn_strides = cfg.MODEL.FCOS.FPN_STRIDES
self.center_sampling_radius = cfg.MODEL.FCOS.CENTER_SAMPLING_RADIUS
self.iou_loss_type = cfg.MODEL.FCOS.IOU_LOSS_TYPE
self.norm_reg_targets = cfg.MODEL.FCOS.NORM_REG_TARGETS
# we make use of IOU Loss for bounding boxes regression,
# but we found that L1 in log scale can yield a similar performance
self.box_reg_loss_func = IOULoss(self.iou_loss_type)
self.centerness_loss_func = nn.BCEWithLogitsLoss(reduction="sum")
def __init__(self, cfg):
self.cls_loss_func = SigmoidFocalLoss(
cfg.MODEL.FCOS.LOSS_GAMMA,
cfg.MODEL.FCOS.LOSS_ALPHA
)
# class_weights = torch.ones(cfg.MODEL.FCOS.NUM_CLASSES).cuda()
# class_weights[1] = 0.4
# self.cls_loss_func_ce = nn.CrossEntropyLoss(reduction="sum")
self.fpn_strides = cfg.MODEL.FCOS.FPN_STRIDES
self.center_sampling_radius = cfg.MODEL.FCOS.CENTER_SAMPLING_RADIUS
self.iou_loss_type = cfg.MODEL.FCOS.IOU_LOSS_TYPE
self.norm_reg_targets = cfg.MODEL.FCOS.NORM_REG_TARGETS
# we make use of IOU Loss for bounding boxes regression,
# but we found that L1 in log scale can yield a similar performance
self.box_reg_loss_func = IOULoss(self.iou_loss_type)
self.f1_loss = F1_Loss(num_classes=cfg.MODEL.FCOS.NUM_CLASSES)
def make_retinanet_loss_evaluator(cfg, box_coder):
matcher = Matcher(
cfg.MODEL.RETINANET.FG_IOU_THRESHOLD,
cfg.MODEL.RETINANET.BG_IOU_THRESHOLD,
allow_low_quality_matches=True,
)
sigmoid_focal_loss = SigmoidFocalLoss(cfg.MODEL.RETINANET.LOSS_GAMMA,
cfg.MODEL.RETINANET.LOSS_ALPHA)
loss_evaluator = RetinaNetLossComputation(
matcher,
box_coder,
generate_retinanet_labels,
sigmoid_focal_loss,
bbox_reg_beta=cfg.MODEL.RETINANET.BBOX_REG_BETA,
regress_norm=cfg.MODEL.RETINANET.BBOX_REG_WEIGHT,
)
return loss_evaluator
def __init__(self, cfg):
self.cls_loss_func = SigmoidFocalLoss(cfg.MODEL.EMBED_MASK.LOSS_GAMMA,
cfg.MODEL.EMBED_MASK.LOSS_ALPHA)
self.fpn_strides = cfg.MODEL.EMBED_MASK.FPN_STRIDES
self.center_on = cfg.MODEL.EMBED_MASK.CENTER_ON
self.center_sampling_radius = cfg.MODEL.EMBED_MASK.CENTER_POS_RADIOS
self.iou_loss_type = cfg.MODEL.EMBED_MASK.IOU_LOSS_TYPE
self.norm_reg_targets = cfg.MODEL.EMBED_MASK.NORM_REG_TARGETS
self.box_reg_loss_func = IOULoss(self.iou_loss_type)
self.centerness_loss_func = nn.BCEWithLogitsLoss(reduction="sum")
########## mask prediction ############
self.mask_loss_func = LovaszHinge(reduction='none')
self.mask_scale_factor = cfg.MODEL.EMBED_MASK.MASK_SCALE_FACTOR
self.object_sizes_of_interest = [
[-1, cfg.MODEL.EMBED_MASK.FPN_INTEREST_SPLIT[0]],
[
cfg.MODEL.EMBED_MASK.FPN_INTEREST_SPLIT[0],
cfg.MODEL.EMBED_MASK.FPN_INTEREST_SPLIT[1]
],
[
cfg.MODEL.EMBED_MASK.FPN_INTEREST_SPLIT[1],
cfg.MODEL.EMBED_MASK.FPN_INTEREST_SPLIT[2]
],
[
cfg.MODEL.EMBED_MASK.FPN_INTEREST_SPLIT[2],
cfg.MODEL.EMBED_MASK.FPN_INTEREST_SPLIT[3]
], [cfg.MODEL.EMBED_MASK.FPN_INTEREST_SPLIT[3], INF]
]
self.sample_in_mask = cfg.MODEL.EMBED_MASK.SAMPLE_IN_MASK
self.sample_pos_iou_th = cfg.MODEL.EMBED_MASK.SAMPLE_POS_IOU_TH
self.box_padding = cfg.MODEL.EMBED_MASK.BOX_PADDING
self.fix_margin = cfg.MODEL.EMBED_MASK.FIX_MARGIN
prior_margin = cfg.MODEL.EMBED_MASK.PRIOR_MARGIN
self.init_margin = -math.log(0.5) / (prior_margin**2)
self.loss_mask_alpha = cfg.MODEL.EMBED_MASK.LOSS_MASK_ALPHA
self.loss_smooth_alpha = cfg.MODEL.EMBED_MASK.LOSS_SMOOTH_ALPHA