Version 2.0.0 runs slowly

[Anqi00]

When I switched to version 2.0.0, the multi-camera matching issue was resolved by adding parameters. general: camera_model: "opencv" single_camera: false

0: DJI Drone

cam0: camera_model: "opencv" intrinsics: ~ images: "*.JPG"

1: Azure Kinect

cam1: camera_model: "opencv" intrinsics: [ 894.637116, 894.415775, 943.383850, 563.671383, 0.078929, -0.045787, 0.003979, -0.002581 ] images: "*.png"

However, I encountered a new problem when I ran python -m deep_image_matching --dir assets/orchard --pipeline superpoint+lightglue --camera_options config/my_cameras.yaml --tiling When running the preselection command, it runs very, very slowly, as if stuck, after displaying "Deep Image Matching loaded in 2.728 seconds." I used all the default settings. But this doesn't happen when using the old version. What's the reason?

[lcmrl]

Good that the camera loading is solved. For the new issue, please could you post the complete command that you have run please? In alternative I also suggest to run the code in the new proposed way running ./demo.py, just properly changing:

args = {
    "dir": "./assets/example_cyprus",
    "pipeline": "superpoint+lightglue",
    "strategy": "bruteforce",
    "quality": "medium",
    "tiling": "none",
    "camera_options": "./assets/example_cyprus/cameras.yaml",
    "openmvg": None,
    "force": True,  # Remove existing features and matches
}

Let me know if you still encounter the issue

[Anqi00]

When I modify the args content as follows: args = { "dir": "./assets/orchard2", "pipeline": "superpoint+lightglue", "strategy": "bruteforce", "quality": "medium", "tiling": "none", "camera_options": "./config/my_cameras.yaml", "openmvg": None, "force": True, # Remove existing features and matches } The output after 10 minutes is still the same:

(deep-image-matching) anqi@anqi-LOQ-16IRH8:~/code/2D3D_macting/auto_matcing_framwork_integrate/deep-image-matching$ python demo.py /home/anqi/code/2D3D_macting/auto_matcing_framwork_integrate/deep-image-matching/src/deep_image_matching/thirdparty/LightGlue/lightglue/lightglue.py:24: FutureWarning: torch.cuda.amp.custom_fwd(args...) is deprecated. Please use torch.amp.custom_fwd(args..., device_type='cuda') instead. @torch.cuda.amp.custom_fwd(cast_inputs=torch.float32) 2025-08-29 21:41:54 | [WARNING ] Not possible to import tkinter Deep Image Matching loaded in 2.642 seconds.

And the same is true when running the example data.

[lcmrl]

With the same input parameters, what is the difference in execution time between versions 1.3 and 2.0?

[Anqi00]

In version 2.0 or the dev branch of version 1.3, for example, the first run goes smoothly, taking about 1 minute. The second run of the same command and the same dataset takes 10 minutes or more, and the third run may take 1 minute or 30 minutes.

[Anqi00]

"Deep Image Matching loaded in 2.715 seconds. PNG file does not have exif data." Sometimes there's a long delay between these two output lines. Is this process extracting the camera's intrinsic parameters?

[lcmrl]

Ok thanks for reporting. Are you on linux or windows? How many images are you processing and which is the resolution? I also had some delays but never so long, I'll look into it. Clearly if you also want to have a look feel free =)

[Anqi00]

I am using Linux, and there are 6 pictures in total, five of which are 4k resolution and one is 2k resolution.

[lcmrl]

Hi, I was not able to replicate the issue, do you still have this issue? Or did you solve in any way?