Ruizhi Shao

We run our test successfully with taichi==0.7.15, so you can try to install taichi==0.7.15.

Hi, Thanks for your advice. We take the last level feature since we only have one attention module. It may be more efficient to build several attention modules on features...

Hi! To resolve your issue, we clarify our camera model and provide some examples in `CUSTOM.md`. As for training, we train our diffusion model with perspective projection.

Hi, In our implementation, scale_mat describes the center and size of the dynamic object. For example, If the bounding box of the object has a size of [2,3,4] and its...

Hi! I have tested several scenarios using only sparse views (

Hi, Tensor4D renderer will return `mid_z_vals` and `weights`. You can sum them to obtain z values. Also, please remember that if your scale_mat is not an identity matrix, this z...

Hi @RuoyiDu, It requires ~15GB of memory in most cases. You can also decrease batch_size or use fp16 mode to further reduce memory usage.

Sorry for the delayed response, as I was busy with another project. "world_mat" can be considered as the projection matrix. It is obtained by multiplying the intrinsic and extrinsic camera...