def _make_detr(
backbone_name: str,
num_queries=100,
mask=False,
qa_dataset=None,
predict_final=False,
text_encoder="roberta-base",
contrastive_align_loss=True,
):
hidden_dim = 256
backbone = _make_backbone(backbone_name, mask)
transformer = Transformer(d_model=hidden_dim,
return_intermediate_dec=True,
text_encoder_type=text_encoder)
detr = MDETR(
backbone,
transformer,
num_classes=255,
num_queries=num_queries,
qa_dataset=qa_dataset,
predict_final=predict_final,
contrastive_align_loss=contrastive_align_loss,
contrastive_hdim=64,
)
if mask:
return DETRsegm(detr)
return detr
def _make_detr(backbone_name: str, dilation=False, num_classes=91, mask=False):
hidden_dim = 256
backbone = Backbone(backbone_name, train_backbone=True, return_interm_layers=mask, dilation=dilation)
pos_enc = PositionEmbeddingSine(hidden_dim // 2, normalize=True)
backbone_with_pos_enc = Joiner(backbone, pos_enc)
backbone_with_pos_enc.num_channels = backbone.num_channels
transformer = Transformer(d_model=hidden_dim, return_intermediate_dec=True)
detr = DETR(backbone_with_pos_enc, transformer, num_classes=num_classes, num_queries=100)
if mask:
return DETRsegm(detr)
return detr
def build(args):
num_classes = 20 if args.dataset_file != 'coco' else 91
if args.dataset_file == "coco_panoptic":
num_classes = 250
device = torch.device(args.device)
backbone = build_backbone(args)
transformer = build_transformer(args)
model = DETR(
args,
backbone,
transformer,
num_classes=num_classes,
num_queries=args.num_queries,
aux_loss=args.aux_loss,
)
if args.masks:
model = DETRsegm(model, freeze_detr=(args.frozen_weights is not None))
matcher = build_matcher(args)
weight_dict = {'loss_ce': 1, 'loss_bbox': args.bbox_loss_coef}
weight_dict['loss_giou'] = args.giou_loss_coef
if args.masks:
weight_dict["loss_mask"] = args.mask_loss_coef
weight_dict["loss_dice"] = args.dice_loss_coef
# TODO this is a hack
if args.aux_loss:
aux_weight_dict = {}
for i in range(args.dec_layers - 1):
aux_weight_dict.update(
{k + f'_{i}': v
for k, v in weight_dict.items()})
weight_dict.update(aux_weight_dict)
losses = ['labels', 'boxes', 'cardinality']
if args.masks:
losses += ["masks"]
criterion = SetCriterion(num_classes,
matcher=matcher,
weight_dict=weight_dict,
eos_coef=args.eos_coef,
losses=losses)
criterion.to(device)
postprocessors = {'bbox': PostProcess()}
if args.masks:
postprocessors['segm'] = PostProcessSegm()
if args.dataset_file == "coco_panoptic":
is_thing_map = {i: i <= 90 for i in range(201)}
postprocessors["panoptic"] = PostProcessPanoptic(is_thing_map,
threshold=0.85)
return model, criterion, postprocessors
def build(args):
# the `num_classes` naming here is somewhat misleading.
# it indeed corresponds to `max_obj_id + 1`, where max_obj_id
# is the maximum id for a class in your dataset. For example,
# COCO has a max_obj_id of 90, so we pass `num_classes` to be 91.
# As another example, for a dataset that has a single class with id 1,
# you should pass `num_classes` to be 2 (max_obj_id + 1).
# For more details on this, check the following discussion
# https://github.com/facebookresearch/detr/issues/108#issuecomment-650269223
num_classes = 20 if args.dataset_file != 'coco' else 91
if args.dataset_file == "coco_panoptic":
# for panoptic, we just add a num_classes that is large enough to hold
# max_obj_id + 1, but the exact value doesn't really matter
num_classes = 250
device = torch.device(args.device)
backbone = build_backbone(args)
if int(os.environ.get("cross_transformer", 0)):
transformer = build_cross_transformer(args)
elif int(os.environ.get("sparse_transformer", 0)):
transformer = build_sparse_transformer(args)
elif int(os.environ.get("linear_transformer", 0)):
transformer = build_linear_transformer(args)
else:
transformer = build_transformer(args)
model = DETR(
backbone,
transformer,
num_classes=num_classes,
num_queries=args.num_queries,
aux_loss=args.aux_loss,
)
if args.masks:
model = DETRsegm(model, freeze_detr=(args.frozen_weights is not None))
matcher = build_matcher(args)
weight_dict = {'loss_ce': 1, 'loss_bbox': args.bbox_loss_coef}
weight_dict['loss_giou'] = args.giou_loss_coef
if args.masks:
weight_dict["loss_mask"] = args.mask_loss_coef
weight_dict["loss_dice"] = args.dice_loss_coef
# TODO this is a hack
if args.aux_loss:
aux_weight_dict = {}
for i in range(args.dec_layers - 1):
aux_weight_dict.update(
{k + f'_{i}': v
for k, v in weight_dict.items()})
weight_dict.update(aux_weight_dict)
losses = ['labels', 'boxes', 'cardinality']
if args.masks:
losses += ["masks"]
criterion = SetCriterion(num_classes,
matcher=matcher,
weight_dict=weight_dict,
eos_coef=args.eos_coef,
losses=losses)
criterion.to(device)
postprocessors = {'bbox': PostProcess()}
if args.masks:
postprocessors['segm'] = PostProcessSegm()
if args.dataset_file == "coco_panoptic":
is_thing_map = {i: i <= 90 for i in range(201)}
postprocessors["panoptic"] = PostProcessPanoptic(is_thing_map,
threshold=0.85)
return model, criterion, postprocessors
def __init__(self, cfg):
super().__init__()
self.device = torch.device(cfg.MODEL.DEVICE)
self.num_classes = cfg.MODEL.DETR.NUM_CLASSES
self.mask_on = cfg.MODEL.MASK_ON
hidden_dim = cfg.MODEL.DETR.HIDDEN_DIM
num_queries = cfg.MODEL.DETR.NUM_OBJECT_QUERIES
# Transformer parameters:
nheads = cfg.MODEL.DETR.NHEADS
dropout = cfg.MODEL.DETR.DROPOUT
dim_feedforward = cfg.MODEL.DETR.DIM_FEEDFORWARD
enc_layers = cfg.MODEL.DETR.ENC_LAYERS
dec_layers = cfg.MODEL.DETR.DEC_LAYERS
pre_norm = cfg.MODEL.DETR.PRE_NORM
# Loss parameters:
giou_weight = cfg.MODEL.DETR.GIOU_WEIGHT
l1_weight = cfg.MODEL.DETR.L1_WEIGHT
deep_supervision = cfg.MODEL.DETR.DEEP_SUPERVISION
no_object_weight = cfg.MODEL.DETR.NO_OBJECT_WEIGHT
N_steps = hidden_dim // 2
d2_backbone = MaskedBackbone(cfg)
backbone = Joiner(d2_backbone,
PositionEmbeddingSine(N_steps, normalize=True))
backbone.num_channels = d2_backbone.num_channels
transformer = Transformer(
d_model=hidden_dim,
dropout=dropout,
nhead=nheads,
dim_feedforward=dim_feedforward,
num_encoder_layers=enc_layers,
num_decoder_layers=dec_layers,
normalize_before=pre_norm,
return_intermediate_dec=deep_supervision,
)
self.detr = DETR(backbone,
transformer,
num_classes=self.num_classes,
num_queries=num_queries,
aux_loss=deep_supervision)
if self.mask_on:
frozen_weights = cfg.MODEL.DETR.FROZEN_WEIGHTS
if frozen_weights != '':
print("LOAD pre-trained weights")
weight = torch.load(
frozen_weights,
map_location=lambda storage, loc: storage)['model']
new_weight = {}
for k, v in weight.items():
if 'detr.' in k:
new_weight[k.replace('detr.', '')] = v
else:
print(f"Skipping loading weight {k} from frozen model")
del weight
self.detr.load_state_dict(new_weight)
del new_weight
self.detr = DETRsegm(self.detr, freeze_detr=(frozen_weights != ''))
self.seg_postprocess = PostProcessSegm
self.detr.to(self.device)
# building criterion
matcher = HungarianMatcher(cost_class=1,
cost_bbox=l1_weight,
cost_giou=giou_weight)
weight_dict = {"loss_ce": 1, "loss_bbox": l1_weight}
weight_dict["loss_giou"] = giou_weight
if deep_supervision:
aux_weight_dict = {}
for i in range(dec_layers - 1):
aux_weight_dict.update(
{k + f"_{i}": v
for k, v in weight_dict.items()})
weight_dict.update(aux_weight_dict)
losses = ["labels", "boxes", "cardinality"]
if self.mask_on:
losses += ["masks"]
self.criterion = SetCriterion(
self.num_classes,
matcher=matcher,
weight_dict=weight_dict,
eos_coef=no_object_weight,
losses=losses,
)
self.criterion.to(self.device)
pixel_mean = torch.Tensor(cfg.MODEL.PIXEL_MEAN).to(self.device).view(
3, 1, 1)
pixel_std = torch.Tensor(cfg.MODEL.PIXEL_STD).to(self.device).view(
3, 1, 1)
self.normalizer = lambda x: (x - pixel_mean) / pixel_std
self.to(self.device)