🤔 学习，反思，与胡思乱想
中英日混杂，请谅解。Posts are in English, Chinese and Japanese. Contact:
Paper Reading
2023-09-14
3️⃣ DreamFusion: Text-to-3D using 2D Diffusion
"DreamFusion" proposes Score Distillation Sampling (SDS) which optimizes a randomly-initialized 3D model (NeRF) via gradient descent such that its 2D renderings from random angles achieve a low loss.
2023-09-14
3️⃣ MVDream: Multi-view Diffusion for 3D Generation
“MVDream” proposes a multi-view diffusion model as a prior for 3D generation trained on a multi-view dataset and pre-trained image diffusion models, and uses Score Distillation Sampling to solve the 3D consistency problem.
2023-09-14
🎞️ Text2Video-Zero: Text-to-Image Diffusion Models are Zero-Shot Video Generators
"Text2Video-Zero" uses a modified Stable Diffusion model with cross-frame attention and motion dynamics to generate videos without training or optimization.