Chenyu

Thank you~

Check the dataset from this link: https://drive.google.com/drive/folders/128yBriW1IG_3NJ5Rp7APSTZsJqdJdfc1

It seems you should implement the logic by yourself. Trying to connect to the server every 1 or 2 seconds until the server established.

I refer to [the line](https://github.com/ashawkey/torch-ngp/blob/c68e05d0be7b9b1b2591320cff988467ea38253e/nerf/network_tcnn.py#L31). Sorry that I cannot show the full command, because I use the nerf part of this repo as part of my code base and I...

Besides, are you available to provide the implementation of the backward pass for the `near_far_from_aabb` function in `raymarching.cu`? Suppose we want to optimize camera poses, then it's necessary to compute...

Sure, the relative pose computation is efficient. But we still have solid reasons to store the relative poses: (1) Sometimes I want to make a comparison with global SfM methods....

I have this problem with torch 1.11 and cuda 11.3 while I use NVIDIA 4090. Tracing stacks: ```sh File "/home/xxx/anaconda3/envs/conerf/lib/python3.9/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context return func(*args, **kwargs) File "/media/.../model/render_utils.py", line...

I'm here to contribute more data points. I encountered the same issue with torch 1.11 and cuda 11.3 while I use NVIDIA 4090. The issue always happens for the **render_image()**...

I'm also curious about how the combined sampling is implemented. It is not clear whether the proposal net uses the transmittance obtained from the occupancy grid sampler for importance sampling,...

How do you build the database, by COLMAP or by DAGSfM? Since DAGSfM stores relative camera poses instead of the essential/fundamental/homography matrices, the database built by COLMAP is not compatible...