easyvolcap: dealing with scene scale

configs/datasets/neural3dv/neural3dv.yaml

@@ -3,7 +3,7 @@ dataloader_cfg:
         ratio: 0.5 # scale images down during loading (thus affects rendering and testing)
         near: 4.0 # camera near far range, affects ray sampling and some cost volume based algorithms
         far: 60.0 # camera near far range, affects ray sampling and some cost volume based algorithms
-        scene_scale: 13.3333333
+        scene_scale: 9.0 # might be wrong, need accurate scale
         bounds: [[-20, -15, 4.0], [20, 15, 25]] # axis-aligned bounding box of the scene, affects ray sampling and bounding box of some explicit algorithms
         view_sample: [0, 21, 1] # these can be omitted if the dataset is present (defaults to [0, null, 1], will inference frame count from number of images)
         frame_sample: [0, 300, 1] # these can be omitted if the dataset is present (defaults to [0, null, 1], will inference frame count from number of cameras)
@@ -14,7 +14,7 @@ val_dataloader_cfg:
         focal_ratio: 0.5 # when generating completely novel views (volumetric_video_inference_dataset), use a smaller focal length to make the scene appear larger
         frame_sample: [0, 300, 100] # make rendering faster, controls which of the views and frames to load
     sampler_cfg:
-        view_sample: [0, 21, 10] # make rendering faster, if doing IBR, change the view_sample in sampler_cfg to control which view to render
+        view_sample: [0, 21, 1] # make rendering faster, if doing IBR, change the view_sample in sampler_cfg to control which view to render
 
 viewer_cfg:
     camera_cfg:

scripts/gaussian/convert_fdgs_pcd.py

@@ -0,0 +1,22 @@
+"""
+Convert fdgs npz to standard 4d gaussian ply
+"""
+import numpy as np
+from easyvolcap.utils.console_utils import *
+from easyvolcap.utils.fdgs_utils import save_ply
+
+
+@catch_throw
+def main():
+    args = dotdict(
+        input='data/trained_model/fdgs_flame_salmon/latest.npz',
+        output='data/result/fdgs_flame_salmon/POINT/latest.ply',
+    )
+    args = dotdict(vars(build_parser(args, description=__doc__).parse_args()))
+    npz = np.load(args.input)
+    save_ply(npz, args.output)
+    log(yellow(f'Saved converted PLY to: {blue(args.output)}'))
+
+
+if __name__ == '__main__':
+    main()