Practical-RIFE
Practical-RIFE copied to clipboard
hzwer
We are developing more practical approach for users based on RIFE.
Hey, I see that in inference image by default you use powers of 2 for interpolation, but also support passing arbitrary "ratio". I'm trying to understand how RIFE is being...
Hello everyone Like the rest of the community who follow your work, we can't wait to see the new version of Rife. As for many I use SVP 4 Pro...
This PR adds PyTorch's [mps (Metal Performance Shaders)](https://pytorch.org/docs/stable/notes/mps.html) backend support on macOS systems, which improves performance from ~1 it/s (pure CPU) to ~7 it/s on Apple M4 chips. Note that...
My expectation on the --montage flag would be to see the original untouched video on (I'm guessing the left side), but, regardless of side, both images in the video have...
作者你好，感谢你为社区所做的开源工作，关于在训练插帧模型的损失函数方面 有些问题想向你请教。 我目前采用的模型也是类似RIFE这种无预训练光流模块的插帧模型，实验发现： 1）用Laplacian Loss会使模型在小物体细节上建模得更好（比如滚动的字幕），但容易出现奇怪的光流错误； 2）用L1 Loss + Vgg Loss组合 则在插帧效果上更稳定，很大程度缓解了光流出错的概率，但在小物体细节上会有退化。 我的理解是LapLoss更让模型更关注于高频细节的建模，但忽略了低频信息。请问你怎么看待这两种损失的偏好，以及在Practical-RIFE中抛弃了LapLoss的原因？ 此外，我试图融合这1）2）两种损失的优点（简单的加权融合），但发现模型貌似难以收敛（加权后的PSNR都低于 单独用其中一种损失），感觉两者之间似乎存在某种对抗？想请问你有何建议。
Hey! I interpolated two EXR frames with the inference_img script and when some pixels' brightness values are over 10, artefacts like this form: https://github.com/user-attachments/assets/3961c4fd-7492-4623-84e0-6d6361d36225 High dynamic EXRs are widely used...
关于模型的泛化性能
您好！ @hzwer 我尝试使用了4.25的训练代码从头训练模型，使用的是vimeo_septuplet+ATD12k+adobe240fps的训练集，验证集使用的是vimeo_septuplet的test_list.虽然指标和您在https://github.com/hzwer/ECCV2022-RIFE/issues/293 的类似，但是在验证集以外的场景上做推理效果却很差，远远比不上您发布的4.25预训练模型。 请问您是有其他的数据集，或者有别的训练方法提高了模型的泛化性能了吗？
Dear hzwer, First of all, I want to congratulate you on the amazing work you've done with the RIFE repository. The quality of the video interpolation is impressive and has...
I am looking for Rife 4.6 general model .But I can't find it. i need Rife 4.6 general model. many anime editors said that 4.6 is the best model right...
