LokiXun

Hi, it such a amazing work! I am trying to train the model on another training data, and I am wondering Is there anywhere to get Test, validation set for...

loss explosion when training on custom Dataset

3

Hi, it an awesome work! May I ask some help, I met some problems when training the model on REDS video dataset. When the training elapses about 40K iterations, the...

Hi, I have question on employing DCNv2 on torch2.0.0+CUDA11.7. Is there any DCNv2 repository could support this configuration? If not, what is the resaon for the failed compatibility? And what...

Hi, I am using pretrained model for video denoising. Checking the code, the model split the frames into overlapped patches to process. However, for the result, the first few patches...

Hi, I have a question about the **Why adding extra conv layers in the Encoder**. And this encoder structure is used in the latter SOTA methods like E2FGVI, Pro-painter. In...

Hi, I am wondering why the Prompt-embedding in StableDiffusion extracted from the penultimate layer of CLIP ViT-H/14 text-encoder? Why not using the origin clip feature just like the image feature...

Question in paper: 1. how to get the mask region as input; 2. Any comparison with ProPainter, just curious

1

For Q1. how to get the mask region as input; Maybe it just doing frame-wise decrease with GT and binarize the mask?

Question in paper: For ablation study "Compare ControlNet and LAControlNet"

5

Hi, I have a question in paper. May I ask a question In the ablation study of "Compare ControlNet and LAControlNet". Figure7(c), the w/ ControlNet result is trained from scratch...

Hi, I have a question for visual grounding. I have a 720x1280 image and i want to describe the region in `[0,0, 512,512]` (x1,y1, x2,y2)so I follow the CogVLM1's suggestion...

Hi, It's a work with great impact. And I had a problem to understand the formula in eq52 (Accelerate sampling). May I wondering that: 1. why eq52 have a additional...