DeDoDe

DeDoDe copied to clipboard

As the title.

xushangnjlh

Any reason for dual softmax matcher for inference?

1

Hi, I am wondering, can we just use mutual nearest neighbor matching for speed at inference time? Dual softmax normalizes the scores, but doesn't change the actual neighbors (matches).

ducha-aiki

About the CE loss

1

Great Job! I wonder why the target probability like this? Seems sum of a log prob of the covisible (also depth consistent) area scoremap, and a keypointmap prob here? Why...

xushangnjlh

Hey! I see that you used DinoV2-L [here](https://github.com/Parskatt/DeDoDe/blob/0928debd09de6e814c90647a639a7219437bf49a/DeDoDe/encoder.py#L53C128-L53C129) Have you tried the smaller versions of dinov2? I couldn't find anything about it in the paper.

felipecadar

Thank you for sharing; As stated in the article, DeDoD v2-L+RoMa has the best effect. Can you please advise on how to use this combination

zhengshunkai

Thank you for your outstanding work. How can this project be deployed on Windows?

wuzelei123

question about detector and discriptor

5

Is there a reason why discriptor and detector used VGG19 network architecture? Not using ResNet or other network architecture

ye450450

Possible bug ?

2

Hi, I think I found a small typo (or maybe I just missed something). In the file: https://github.com/Parskatt/DeDoDe/blob/0928debd09de6e814c90647a639a7219437bf49a/DeDoDe/matchers/dual_softmax_matcher.py I notice that "DualSoftMaxMatcher" returns batch_inds, shouldn't it return inds?

fratopa

Unsure how to use the S-Version

3

Hello, Im not sure how to use the published S weights. There only seems to be a `dedode_detector_L` and `dedode_detector_B` not S. Neither version work with the S weights. How...

geoeo

Could you release small/tiny/nano version of detector and descriptor?

12

Hello, thank you for your amazing job! I'm really interesting of your job and want to deploy DeDoDe on mobile devices(laptop, even CPU) for some self-driving works. But I find...

zhongqiu1245