Official implementation of “GaussianTalker: Real-Time High-Fidelity Talking Head Synthesis with Audio-Driven 3D Gaussian Splatting” by Kyusun Cho, Joungbin Lee, Heeji Yoon, Yeobin Hong, Jaehoon Ko, Sangjun Ahn and Seungryong Kim
I feel quite confused about the so-called real-time inference or rendering. just like a scene Obama.mp4 in the picture above, it has cost me 1hours and more on executing "data_utils/process.py",and it still doesnt finish with 200 steps left of "fine frame-wise fitting". My machine is 1070Ti 8GB with 24CPUs.
If add the training and rendering, I guess the result will outcome in half a day at least.
Of course, the rendering quality and speed of 3Dgs is amazing. so it is a excellent job what the author has done. But how can i realize real-time on any video? who can help me? Must I do process and train for every custom video?
I feel quite confused about the so-called real-time inference or rendering. just like a scene Obama.mp4 in the picture above, it has cost me 1hours and more on executing "data_utils/process.py",and it still doesnt finish with 200 steps left of "fine frame-wise fitting". My machine is 1070Ti 8GB with 24CPUs.
If add the training and rendering, I guess the result will outcome in half a day at least.
Of course, the rendering quality and speed of 3Dgs is amazing. so it is a excellent job what the author has done. But how can i realize real-time on any video? who can help me? Must I do process and train for every custom video?