search

cvlab-kaist / DiffMatch

Official implementation of "Diffusion Model for Dense Matching" (ICLR'24 Oral)
154 stars 6 forks source link

test on our own data #5

Open zgdjcls opened 6 months ago

zgdjcls commented 6 months ago

Hi, I'm shocked by the performance of DiffMatch model and interested in how well it could be on specialized data. I want to know do you have any plan to release the code that could run on our own data(i.e. an images pair). Thank you for any info updating.

SuhZhang commented 5 months ago

Hi, I encounter a similar problem when running DiffMatch on public datasets such as MegaDepth and Scannet. It seems that in the test codes the value 'mask' is indispensable which comes from the ground truth corresponding relationship, perplexing me whether to access promising performance without 'mask' for it is unknown while testing. Have you accomplished the implementation on your data? How to make it available? Thanks for your kind proposals.

luocha0107 commented 4 months ago

我也想知道，麻烦楼一楼二看到回复一下，谢谢你嘿嘿。 I so want to know.

boilcy commented 4 months ago

This project borrow code from Dense Matching. It says how to prepare the dataset. https://github.com/PruneTruong/DenseMatching?tab=readme-ov-file#training-datasets-downloading-

看这个库的介绍，里面有说怎么处理数据集得到image_pairs和flows。

Fine, this repo still use a magic csv file... 这个库的csv也没说哪来的， DEPD数据集自带吗？我还没下

here https://github.com/PruneTruong/DenseMatching/issues/13

好的是写了...

lichenyi112 commented 3 months ago

Hi, I'm shocked by the performance of DiffMatch model and interested in how well it could be on specialized data. I want to know do you have any plan to release the code that could run on our own data(i.e. an images pair). Thank you for any info updating.

have you succeeded to use DiffMatch to match your own image pairs?