If you want to test your own data, you should refer to the following steps:
-
Select about 300-600 images from video. Set the path of video and run process_video.py to get 300-600 frames.
process_video.py -
Download COLMAP.exe in Windows (also can be other platforms), run COLMAP.exe using the above images and export the model with .txt format. Just same with our given examples.
-
Run get transoform.json by running submodules/instant-ngp/scripts/colmap2nerf.py. You can also refer to the instruction of Instant-NGP. It is the same steps.
data-folder$ python [path-to-instant-ngp]/scripts/colmap2nerf.py --colmap_matcher exhaustive --aabb_scale 4 -
You must download Instant-NGP.exe in Windows, drag the colmap folder (contain images folder and transform.json) to the instant-ngp program. After training 2-3 minutes, crop the bbox to clear up the background and then save the pretrained weight and add the key frame in front view.
- Run instant-ngp.exe
- Crop the bbox size to clear the background and save the pretrained weight. You will get "base.ingp" in "colmap folder"
- Change the path as "key_frame.json" in Camera path tool.
- Select a suitable front view as the key frame (We will create 16 fixed views base on it when using DeepMVSHair). Click "Add from cam" to add a key frame and click "Save" to save its camera parameters. Then you will get "key_frame.json"
After the above steps, you should have colmap_text folder, transofrm.json, base.ingp, key_frame.json. you can continue the process in the main page.