Error: Unexpected key(s) in state_dict: "transformer.text_encoder.embeddings.position_ids"

[bernard0047]

I get this error when I run:

python inference/main.py -m mdef_detr -i ../../images -c ckpt/MDef_DETR_r101_epoch20.pth -tq "all objects"

[KemingWu]

me too. Do you solve it now?

[qwertfyghjbn]

`def _make_mdef_detr(checkpoints_path):
backbone = _make_backbone()
transformer = MDefDETRTransformer(d_model=256, return_intermediate_dec=False, num_feature_levels=4,
dim_feedforward=1024, text_encoder_type="roberta-base")
model = MDefDETR(backbone=backbone, transformer=transformer, num_classes=255, num_queries=300,
num_feature_levels=4)
checkpoint = torch.load(checkpoints_path, map_location="cpu")

state_dict = checkpoint["model"]
# 移除不兼容的键
if "transformer.text_encoder.embeddings.position_ids" in state_dict:
    del state_dict["transformer.text_encoder.embeddings.position_ids"]
model.load_state_dict(state_dict)

#model.load_state_dict(checkpoint["model"])

return model`

修改mvits_for_class_agnostic_od-main/models/model.py中的_make_mdef_detr函数