Zipei Ma
@yuedajiong > 假设比如第1个时刻点，右手在头顶，假设有一个高斯点表示了右手食指；第n个时刻点，右手挥舞到腹部了，这个时候，右手食指，还是同一个高斯点，通过xyz经过n-1步移动过去，来表示的吗？ 根据论文和代码，高斯的xyz是保持不变的。如果我没理解错的话：在时刻1，头顶部的右手是由头顶部的高斯表示（此时那块高斯的时间t处于峰值）；而在第n时刻，是由腹部的高斯来表示的（同样，那块地方的高斯的时间t处于峰值）
请问有后续的结果吗，是采用 lora 还是直接改 官方提供的 ckpt？
I have tested, it can improve psnr for 0.1-0.2, my config: ``` # Color use_color_correction: true color_mode: "sensor" # If set to 'image', learn separate embedding for each image. If...
Currently, I have down a version. Hope to help guys: ```python import os import torch import numpy as np from detectron2.engine import DefaultTrainer, default_argument_parser, default_setup, launch from detectron2.config import get_cfg...
Is there any bugs? The prediction matches two different cars for one idx in the whole sequence, here's results: [mp4.zip](https://github.com/user-attachments/files/19836242/mp4.zip)
Hi @NanoCode012 , I have updated my deepspeed to the newest version and I still met the bug: ```bash [rank0]: Traceback (most recent call last): [rank0]: File "/inspire/ssd/project/robot3d/mazipei-253107140027/WorldActionModel/main.py", line 69,...
@NanoCode012 Sorry, I think it's a bug from the accelerate.
Have you found it?
Since you're using COLMAP for feature extraction, you don't need the GUI. You can force COLMAP to run in a non-GUI mode by setting the following environment variable before running...
Hi, have you dealt with it? @qiqiApink