search

RaymondWang987 / NVDS

The official repository of the ICCV2023 paper "Neural Video Depth Stabilizer" (NVDS).
MIT License
460 stars 24 forks source link

The stabilizers of this project can be used for other visual tasks, such as video matting #18

Open zhanghongyong123456 opened 10 months ago

zhanghongyong123456 commented 10 months ago

Thank you for a great project，Whether it can be used for other visual tasks, video matting, video optical flow estimation stability

RaymondWang987 commented 10 months ago

Thank you for a great project，Whether it can be used for other visual tasks, video matting, video optical flow estimation stability

Hi! Thanks for your attention to our work. We have already made some extensions of NVDS to other video tasks, e.g., video semantic segmentation. The results also show obvious improvements in consistency than the single-image segmenters, which prove the generality of our framework. You can also try on other tasks such as video matting. The extensions of our framework will be released in the journal version of our paper in several months.

zhanghongyong123456 commented 10 months ago

Thank you for a great project，Whether it can be used for other visual tasks, video matting, video optical flow estimation stability

Hi! Thanks for your attention to our work. We have already made some extensions of NVDS to other video tasks, e.g., video semantic segmentation. The results also show obvious improvements in consistency than the single-image segmenters, which prove the generality of our framework. You can also try on other tasks such as video matting. The extensions of our framework will be released in the journal version of our paper in several months.

Thank you very much for your prompt reply，I didn't find the training code available, so how can I train myself to pick the stabilizer in the video matting， In addition, when can the real-time model plan be tested（Light-weight NVDS-Tiny model） Thanks again

RaymondWang987 commented 9 months ago

Thank you for a great project，Whether it can be used for other visual tasks, video matting, video optical flow estimation stability

Hi! Thanks for your attention to our work. We have already made some extensions of NVDS to other video tasks, e.g., video semantic segmentation. The results also show obvious improvements in consistency than the single-image segmenters, which prove the generality of our framework. You can also try on other tasks such as video matting. The extensions of our framework will be released in the journal version of our paper in several months.

Thank you very much for your prompt reply，I didn't find the training code available, so how can I train myself to pick the stabilizer in the video matting， In addition, when can the real-time model plan be tested（Light-weight NVDS-Tiny model） Thanks again

We currently have no plan to release our training code. If you need some training scripts, you can contact me by email and I can send you a raw version of our finetuning code on NYUDV2.

The real-time NVDS-Small model will be released when the journal version of our paper is about to be ready. Please stay tuned.