LoFTR

LoFTR copied to clipboard

```data['mask0']``` and ```data['mask1']```, indicating the padded regions, are coarse-level as presented in ```megadepth.py.``` while those masks are used to mask out the original-level matching points [here](https://github.com/zju3dv/LoFTR/blob/df7ca80f917334b94cfbe32cc2901e09a80e70a8/src/loftr/utils/supervision.py#L56). Even in case where...

HJoonKwon

Does the size of the input image have to be an integer multiple of 64?

LooperzZ

Reproducing the training results on a megadepth dataset

23

Thank you very much for your excellent work. I recently reproduce the training results on 4 3090 GPUs for 30 epochs based on README. The batch size each GPU is...

FlyFish-space

How to train in MegaDepth without D2-Net data

10

Hi authors, thanks for your great works! I see the F&Q, since we cannot reach D2_net pre-processing data anymore, you suggest us to use MegaDepth directly. May I ask how...

QiuhangLiu

Use of masks in feature detection

3

Is the purpose of masks to provide padding at the border of the image? https://github.com/zju3dv/LoFTR/blob/b4ee7eb0359d0062e794c99f73e27639d7c7ac9f/src/loftr/loftr.py#L35 I have a scenario where I want to detect feature matches in an image except...

sandeepnmenon

在cparse_matching.py文件的第245行，感觉index是反过来的，为什么是 mkpts0_c = torch.stack( [i_ids % data['hw0_c'][1], i_ids // data['hw0_c'][1]], dim=1) * scale0， 而不是 mkpts0_c = torch.stack( [ i_ids // data['hw0_c'][1]，i_ids % data['hw0_c'][1]], dim=1) * scale0

BaochunHe

Megadepth数据集中的深度图解析出来全是0

2

您好，我看源代码在生成gt时会使用megadepth数据集中depth0目录中的.h5文件，我用你们的代码解析.h5文件中的['depth']返回的全是0请问这个怎么解决啊？

18161003959

My goal is to detect key points on two images, and I am wondering if your model can be restricted to a specific area.

Abbsalehi

train and test on PhotoTourism dataset

12

atoshpulatov

Inference time

7

Has anyone tried to get this running using tensorRT or C++? It would be good to know the rough fps for this. Cheers,

YJonmo