@XingyiHe1 The refinement method proposed by MatchAnything, which enables detector-free matches to propagate further, would certainly be beneficial for obtaining matches between distant frames. Anyway, MatchAnything is a nice work. (2/2)

@XingyiHe1 Hi, Xingyi, thank you for your reply. The limitation of detector-free matches in propagating to distant frames is also a phenomenon we observed in our experiments. Therefore, during propagation, we fused multiple methods to try to propagate matches to greater distances. (1/2)

@sierrambonilla @ducha_aiki @zhenjun_zhao Based on the ZEB data on GIM's Github, it seems GIM_DKM would perform far better than GIM_LightGlue. I'm just curious about how the best model, GIM_DKM, would perform. Anyway, mismatched is great. It gives me more about the latest image matching method.

@sierrambonilla @ducha_aiki @zhenjun_zhao Could I ask why GIM_DKM isn't tested in the mismatched paper?