Compare two NeRFs #537
rooi
started this conversation in
Ideas / Feature requests
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Hi,
I would like to compare different snapshots/NeRFs to visualize differences. The visualization is not that important for now, could just be the training visualization starting from a preloaded (different) snapshot. Perhaps it would be nice to have such functionality build in instant-ngp
However, for this to work, the camera poses / coordinate frame should be the same.
I understand colmap has build-in functionality to achieve this, for example using the image_list_path argument
https://colmap.github.io/faq.html#register-localize-new-images-into-an-existing-reconstruction
Unfortunately I cannot get this to work correctly, since the different NeRF models are misaligned. Could someone, perhaps give some pointers to achieve this?
Right now I use:
colmap2nerf.py --run_colmap ...
train using instant_ngp and save the snapshot
on the new images:
colmap feature_extractor --ImageReader.camera_model OPENCV --SiftExtraction.estimate_affine_shape=true --SiftExtraction.domain_size_pooling=true --ImageReader.single_camera 1 --database_path {db} --image_path {images} --image_list_path {args.image_list_path}
colmap mapper --database_path {db} --image_path {images} --image_list_path {args.image_list_path} --output_path {sparse} colmap bundle_adjuster --input_path {sparse}/1 --output_path {sparse}/1 --BundleAdjustment.refine_principal_point 1 colmap model_converter --input_path {sparse}/1 --output_path {text} --output_type TXT
then I run instant-ngp using a snapshot and new training set.
.\build\testbed.exe --scene .\data\nerf\test01 --snapshot .\data\nerf\test07\base1.msgpack --no-train
Thanks!
Beta Was this translation helpful? Give feedback.
All reactions