ryancll

Question about scaling?

1

Hi @zachteed @heilaw @anewell @jiadeng I notice that you scale both the depth map and the translation in pose matrix with scaling ratio 0.1 when training the KITTI dataset. However,...

Text Encoder

6

Thx for the wonderful work! Do you have any plans to support T5 text encoder recently?

Pooled Text Embedding

1

I notice that you use pooled text embedding for T2V models. Using full text embedding for T2I and T2V models should be an good alternative (like officical Latte). Hope this...

https://github.com/wyhsirius/LIA/blob/d120cb29a56e5798b59aa66b975b7ae4f2265140/run_demo.py#L90 Thx for the great work! Could you please explain why we need to use h_start and why you set h_start=None for Ted data?

Scale for rays_o

1

https://github.com/hjxwhy/Block-NeRF/blob/bc16e5cc95ca534b451d4125770bd682569b6776/datasets/memory_dataset.py#L130 Could you please explain why we should scale rays_o by a factor of 100 for waymo dataset?

Thank you for your contributions! In the paper of StylePeople (Sec. 3.2 & supplementary material), you mention that the renderer network can be pretrained on 56 people from AzurePeople dataset....

Considering the maxium length of generated video is limited by GPU memory, I tried splitting a long video into multiple segments. Even though I edit each segment with same prompt...

Negative Prompt

1

Thx for the wonderful work! I'm wondering is that possible to add negative prompt when generating videos for better performance?