Vincentqyw/image-matching-webui

two bugs

fanjing8 opened this issue · 4 comments

fanjing8 commented

问题1：

image-matching-webui/hloc/match_dense.py

Lines 396 to 397 in d97d9af

"keypoints0": kpts0_origin.cpu().numpy(),
"keypoints1": kpts1_origin.cpu().numpy(),

这两行应该是 ：
"keypoints0": kpts0.cpu().numpy(),
"keypoints1": kpts1.cpu().numpy(),

问题2： 缺少 matchers/dedode.py 文件

fanjing8 commented

问题1：

image-matching-webui/hloc/match_dense.py

Lines 396 to 397 in d97d9af

"keypoints0": kpts0_origin.cpu().numpy(),
"keypoints1": kpts1_origin.cpu().numpy(),

这两行应该是 ：
"keypoints0": kpts0.cpu().numpy(),
"keypoints1": kpts1.cpu().numpy(),
问题2： 缺少 matchers/dedode.py 文件

针对问题2：match_dense.py里面 dedode_sparse 用的是 extractors/dedode.py?

Vincentqyw commented

问题1：

image-matching-webui/hloc/match_dense.py

Lines 396 to 397 in d97d9af

"keypoints0": kpts0_origin.cpu().numpy(),
"keypoints1": kpts1_origin.cpu().numpy(),

这两行应该是 ：
"keypoints0": kpts0.cpu().numpy(),
"keypoints1": kpts1.cpu().numpy(),
问题2： 缺少 matchers/dedode.py 文件

  1. Not a bug, please try again and check the visualization. The naming of keypoints may lead to ambiguity. These keypoints are used in the subsequent visualization stage, where they are plotted on the original image. Therefore, resizing the points is necessary. There are no abnormalities in this process.
  2. Key dedode_sparse is not used, so there is no matchers/dedode.py. I will delete the key.

If there are no other issues, please close the question.

fanjing8 commented

问题1：

image-matching-webui/hloc/match_dense.py

Lines 396 to 397 in d97d9af

"keypoints0": kpts0_origin.cpu().numpy(),
"keypoints1": kpts1_origin.cpu().numpy(),

这两行应该是 ：
"keypoints0": kpts0.cpu().numpy(),
"keypoints1": kpts1.cpu().numpy(),
问题2： 缺少 matchers/dedode.py 文件

  1. Not a bug, please try again and check the visualization. The naming of keypoints may lead to ambiguity. These keypoints are used in the subsequent visualization stage, where they are plotted on the original image. Therefore, resizing the points is necessary. There are no abnormalities in this process.
  2. Key dedode_sparse is not used, so there is no matchers/dedode.py. I will delete the key.

If there are no other issues, please close the question.

我是用image0/1和keypoints0/1可视化出来不一致，改之后才会一致。

我理解image0/1和keypoints0/1是对应的，都是scale之后的（带_orig后缀的是scale之前的）。按现在的写法，keypoints0和keypoints0_orig是一样的，都是keypoints0_orig（keypoints1_orig同理）

Vincentqyw commented

I appreciate your insight. To address potential ambiguity, I've enacted some modifications in the subsequent commit:

9162a57#diff-cbb7e4425d46c92eae364a7c62a58ff478af1d596e191b331396de21ab66405fL396-R420

Some Notes:

- Image Detected KeyPoints Matched KeyPoints
wo. scale image0 keypoints0 mkeypoints0
w. scale (view in UI) image0_orig keypoints0 _orig mkeypoints0_orig