Skip to content

Data Processing instructions for MegaDepth Option 1 are incomplete. #3

@SergioRAgostinho

Description

@SergioRAgostinho

Just logging some problems I've discovered.

I've been going over the data preparation instructions for megadepth and noticed that Option 1 is not really an option.

gomatch/tools/README.md

Lines 25 to 31 in 3cd031d

#### Option 1
By default the tool assumes there is a folder with the undistorted SfM reconstruction of MegaDepth symlinked to the `data` folder, as explained in the previous section. If the folder is symlinked, theoretically no arguments need to passed. Just call the tool with
```
$ python tools/prep_megadepth_2D3D_data.py
```
After completion, the tool will generate a number of files in `data/MegaDepth_undistort/data_processed/v2`.

If you follow those instructions it will not generate the necessary file that is then required in the follow-up step

gomatch/tools/README.md

Lines 42 to 49 in 3cd031d

### Generating a Descriptor Cache
Run the tool `tools/extract_features_immatch.py`.
```
python -m tools.extract_features_immatch --immatch_config 'immatch/sift' --dataset 'megadepth'
```
This will generate the descriptor cache for sift.

Our config is looking for

data_file: "sift_pts/megadepth_2d3d_q500ov0.35tp3-10_imsz.irat1.5.sift1024.npy"

but our generation script would never be able to create that file
data_save_path = os.path.join(
save_dir,
f"megadepth_2d3d_q{args.scene_max}ov{args.overlap[0]}-{args.overlap[1]}covis{topk_min}-{topk_max}.npy",
)

the string irat is not even part of the file name construction.

So something got lost during the public release. I'll try to see if I figure out what exactly.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions