demo_colmap output points.py has zero points

[canyou2014]

demo_colmap output points.py has zero points，non error happen, but use demo_gradio run the same datasets, its output look good

[jytime]

Hi it seems the points are filtered out by confidence. Please try to adjust the value here:

https://github.com/facebookresearch/vggt/blob/8e3d2809c1c93fa27298c290605841d7e876b285/demo_colmap.py#L60

[canyou2014]

Yes, it's ok when set conf_thres_value to 1.x, thanks.

[xiemeilong]

It not works even after set conf_thres_value=0.7 vis_thresh 0.1 query_frame_num: 20 max_query_pts 4096.

The same 72 images works in demo_gradio.

python demo_colmap.py --scene_dir /home/xiemeilong/test/labs  --use_ba --shared_camera  --query_frame_num 20  --conf_thres_value=0.7  --vis_thresh 0.1 
Arguments: {'scene_dir': '/home/xiemeilong/test/labs', 'seed': 42, 'use_ba': True, 'max_reproj_error': 16.0, 'shared_camera': True, 'camera_type': 'SIMPLE_PINHOLE', 'vis_thresh': 0.1, 'query_frame_num': 20, 'max_query_pts': 4096, 'fine_tracking': True, 'conf_thres_value': 0.7}
Setting seed as: 42
Using device: cuda
Using dtype: torch.bfloat16
Model loaded
Loaded 72 images from /home/xiemeilong/test/labs/images
Using cache found in /data/ssdc/cache/torch/hub/facebookresearch_dinov2_main
/data/ssdc/cache/torch/hub/facebookresearch_dinov2_main/dinov2/layers/swiglu_ffn.py:51: UserWarning: xFormers is not available (SwiGLU)
  warnings.warn("xFormers is not available (SwiGLU)")
/data/ssdc/cache/torch/hub/facebookresearch_dinov2_main/dinov2/layers/attention.py:33: UserWarning: xFormers is not available (Attention)
  warnings.warn("xFormers is not available (Attention)")
/data/ssdc/cache/torch/hub/facebookresearch_dinov2_main/dinov2/layers/block.py:40: UserWarning: xFormers is not available (Block)
  warnings.warn("xFormers is not available (Block)")
For faster inference, consider disabling fine_tracking
Predicting tracks for query frame 0
Predicting tracks for query frame 21
Predicting tracks for query frame 45
Predicting tracks for query frame 6
Predicting tracks for query frame 1
Predicting tracks for query frame 2
Predicting tracks for query frame 4
Predicting tracks for query frame 8
Predicting tracks for query frame 3
Predicting tracks for query frame 10
Predicting tracks for query frame 5
Predicting tracks for query frame 9
Predicting tracks for query frame 7
Predicting tracks for query frame 11
Predicting tracks for query frame 14
Predicting tracks for query frame 16
Predicting tracks for query frame 12
Predicting tracks for query frame 28
Predicting tracks for query frame 13
Predicting tracks for query frame 37
Not enough inliers per frame, skip BA.
Traceback (most recent call last):
  File "/data/ssdc/sources/vggt/demo_colmap.py", line 298, in <module>
    demo_fn(args)
  File "/data/ssdc/sources/vggt/demo_colmap.py", line 187, in demo_fn
    raise ValueError("No reconstruction can be built with BA")
ValueError: No reconstruction can be built with BA

[jytime]

Hi @xiemeilong , If you are happy with the result of demo_gradio, you can just run demo_colmap without --use_ba. The flag conf_thres_value will only control demo_colmap without BA.