Abnormal image position in model

[0CYH0]

Hello，I attempted to reconstruct several images captured with the same camera. The instructions are as follows:

feature_conf = extract_features.confs['superpoint_aachen']
matcher_conf = match_features.confs['superglue']

pairs_from_exhaustive.main(sfm_pairs, image_list=references_select)
feature_path = extract_features.main(feature_conf, images, outputs, image_list=references_select)
match_path = match_features.main(matcher_conf, sfm_pairs, feature_conf['output'], outputs)

opts = dict(camera_model='SIMPLE_RADIAL')
model = reconstruction.main(sfm_dir, images, sfm_pairs, feature_path, 
                            match_path, camera_mode=pycolmap.CameraMode.SINGLE,
                            image_list=references_select, image_options=opts,
                            verbose=True)

Sometimes there are significant errors in the positions of images reconstructed in the model, as shown in the following picture:

I noticed that the number of points (num_points3D) in this image is significantly fewer compared to the images in normal positions, and the positions seem to be not quite accurate, as shown in the following figures:


Could you please clarify the relationship between 'num_points3D' in the model and the characteristics of the images? How should I address this issue?
Looking forward to your reply.

[sarlinpe]

This is pretty odd. How does the previous/next image in the sequence look? You could maybe visualize the correspondences extracted by SuperGlue and see why there would be so few of them.