Skip to content

Unable to Run MaskAdapter with MAFT+ #9

Open
@DanilaAniva

Description

@DanilaAniva

I am trying to run MaskAdapter with MAFT+, using the configuration file train_semantic_large_eval_a150.yaml. However, I am encountering issues when executing demo.py with MAFT+ weights. The same approach works fine when running MaskAdapter with FC-CLIP, but it fails when switching to MAFT+

What I Am Doing
Downloaded the MAFT+ checkpoint (MAFTP-Large + Mask-Adapter)
gdown --id 12eqDnTYaQlj9QUmWO1Vh9vvB81tKABl5 -O checkpoints/maftp_l.pth
Attempted to run inference with demo.py using the train_semantic_large_eval_a150.yaml

python demo/demo.py \ --config-file configs/mixed-mask-training/maftp/semantic/train_semantic_large_eval_a150.yaml \ --input ../Semantic-Segment-Anything/input_images/image_to_stable_res1.png \ --output output_maft \ --opts MODEL.WEIGHTS ./checkpoints/maftp_l_withadapter.pth
However, this results in an error related to a missing config key:
KeyError: 'Non-existent config key: MODEL.rc_weights'

When i run fcclip with "configs/mixed-mask-training/fc-clip/fcclip/fcclip_convnext_large_eval_ade20k.yaml", it works fine
python demo/demo.py \ --input ../Semantic-Segment-Anything/input_images/*.png ../Semantic-Segment-Anything/input_images/*.jpg \ --output ./output_results \ --opts MODEL.WEIGHTS ./checkpoints/fcclip_maskadapter.pth

Is train_semantic_large_eval_a150.yaml fully compatible with demo.py?
If not, what is the recommended way to run inference for MAFT+?

Is there a different script for running MAFT+ inference instead of demo.py?

demo.py is from HF demo

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions