issue at raymarching

[yyashpatel]

Hi !
I face this error when at the command ray_marching()
Please help me this ... thank you !

File "train.py", line 95, in
main()
File "train.py", line 93, in main
nerf_sim.training()
File "train.py", line 63, in training
self.nerf.render(ray_o, ray_d, rgb_gt, step)
File "/home/yash/Documents/nerf/NeRF/run_NeRF.py", line 64, in render
rgb, acc, depth = self.output.renderer(ray_o,
File "/home/yash/Documents/nerf/NeRF/rendering.py", line 59, in renderer
packed_info, t_start, t_end = self.vol_sampling(ray_o, ray_d, aabb, grid, render_step_size, cone_angle, alpha_thre)
File "/home/yash/Documents/nerf/NeRF/rendering.py", line 46, in vol_sampling
packed_info, t_start, t_end = ray_marching(ray_o,
File "/home/yash/anaconda3/envs/nerf_sim/lib/python3.8/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
return func(*args, **kwargs)
File "/home/yash/anaconda3/envs/nerf_sim/lib/python3.8/site-packages/nerfacc/ray_marching.py", line 156, in ray_marching
grid_roi_aabb = grid.roi_aabb
AttributeError: 'float' object has no attribute 'roi_aabb'

[yyashpatel]

the above issue is resolved, It's working now,

Just one more thing -

the aabb argument that has to be specified in OccupancyGrid(roi_aabb = aabb) is the same as that in ray_marching(scene_aabb=aabb) ?

[liruilong940607]

They can be the same but not necessary to be.

If you have a scene bounded by a bbox, it is recommended to set them to the same.

If you want to use the contraction for the occupancy grid to cover an infinity space, e.g., OccupancyGrid(..., contraction_type=ContractionType.UN_BOUNDED_SPHERE), then the roi_aabb is the region of interest in the infinity space. In this case you shouldn't set the scene_aabb to ray_marching, as it will only march inside this bounding box

[yyashpatel]

Understood thank you !

I just wanted to know , when performing testing , I wonder how the python api commands provided are to be used .

Because in the training part I use occupancy grid in the raymarching() function . And since it gets updated every time, so I am a bit confused about how to test the model once I have save the weights ?

[liruilong940607]

Same way as you do the training — use the occupancy grid as well.

That means you would want to save both your network and the occupancy grid, and reload them when testing

[yyashpatel]

Oh okay , I did get your point.
Saving occupancy grid using torch.save() .?

[liruilong940607]

yeah the occupancy grid is a torch.nn.Module so you can save it in the same way you do with network.

[liruilong940607]

Closed as the problems seem to be all resolved. Feel free to reopen it if not.