match_dense confs on EfficientLoftr

[boyagesmile]

I got bad results when I try to test EfficientLoftr on aachen_v1.1 datasets by using the config:

    "loftr_aachen": {
        "output": "matches-loftr_aachen",
        "model": {"name": "loftr", "weights": "outdoor"},
        "preprocessing": {"grayscale": True, "resize_max": 1024, "dfactor": 8},
        "max_error": 2,  # max error for assigned keypoints (in px)
        "cell_size": 8,  # size of quantization patch (max 1 kp/patch)
    },

I wonder to know how to chagnge "max_error" and "cell_size". Do you have any plan to update the EfficinetLoftr test?

[Inverse-function]

Hello, I also want to test EfficientLoFTR, but I encountered an error where two tensor sizes do not match at the transformer position. How did you solve this? Or could you share your testing script with me? Thank you.

[boyagesmile]

May be you should try to change "dfactor": 8 to "dfactor": 32

Hello, I also want to test EfficientLoFTR, but I encountered an error where two tensor sizes do not match at the transformer position. How did you solve this? Or could you share your testing script with me? Thank you.

[Inverse-function]

Thank you for your reply. Could you please explain what the "dfactor" parameter represents above?