PMRF - 高质量人像超清修复、人脸逼真高清修复 本地一键整合包下载

无言***

PMRF - 高质量人像超清修复、人脸逼真高清修复 本地一键整合包下载

PMRF（Posterior-Mean Rectified Flow） 是一种全新的图像修复算法，旨在实现高质量的图像恢复。
在图像处理技术日新月异的今天，一项名为PMRF（后验均值修正流）的创新算法正在引起业界广泛关注。这项技术巧妙地解决了图像恢复过程中长期存在的失真与感知质量之间的矛盾，为高质量图像重建开辟了新的可能性。

和以往的修复算法不同，PMRF具备高质量修复图像的同时，降低图像失真，还原更加自然逼真度主要能力。特点总结来说：高度保留源图细节的同时做超清修复。


主要功能
图像恢复：处理去噪、超分辨率、盲图像恢复和图像修复等任务，生成自然逼真的图像。例如：
               去噪：去除图像中的噪声，使其更加清晰。
               超分辨率：提高低分辨率图像的细节，使其更接近高分辨率图像。
               修复：填补图像中的缺失部分，例如修复受损的区域或在图像中添加缺失的细节。
颜色恢复：恢复或增强图像的颜色，使其更加自然或符合真实场景。
降低图像失真（MSE）：PMRF通过后验均值预测实现图像恢复，最小化图像的均方误差（MSE），确保生成的图像与真实图像在数值上尽可能接近，失真尽可能小。也就是它不仅关注图片的清晰度，还会确保图片看起来像真实世界中的图像
提高感知质量：PMRF不仅仅追求数值精度，还能够通过校正流模型（Rectified Flow）确保复原图像的感知质量与真实图像保持一致。这意味着，PMRF能够生成视觉上更加逼真的图像，使其在人眼看来与原始图像几乎无异。
处理复杂的图像退化问题：PMRF能够应对各种复杂的图像退化情况，包括噪声、模糊、分辨率降低、颜色丢失等问题，生成视觉质量高且符合真实图像分布的恢复图像。不论图片内容有多复杂，比如细节丰富的面部图像，还是受到多重损坏的图片，PMRF都能很好地处理，并提供优质的修复结果。
优化的图像生成流程：PMRF结合后验均值预测和修正流模型，通过求解常微分方程（ODE）对图像进行“运输”，使得生成的图像既低失真又高质量。它通过在图像分布之间实现最佳映射，达到感知和失真之间的平衡。



使用教程：CPU和独显都可以使用，建议使用4G显存起的N卡，处理速度更快。
操作很简单，上传一张需要修复的照片，参数可以默认，点提交即可。
高级参数设置
随机化种子：这个参数建议开启，如果生成的效果不满意，可以重新提交生成；
输入是一个对齐的方形人脸图像：如果你的源图是一张正方形的，且面部朝向前方，建议勾选；
缩放系数：适用于非正脸的源图素材，可以适当调大该参数；

推理步骤数：数字越大，图像质量越好，同时速度越慢。


下载地址：
https://pan.quark.cn/s/dcf2905a6560
https://www.123pan.com/s/OYeA-gToBh

jiao***

启动正常，点运行后报错，：“CUDA runtime error: no kernel image is available for execution on the device”
-----------------------------------------------------------------------------------------------------------------------------

-----------------------------------------------------
更多实用好玩的AI应用，访问
软件 WebUI正在启动中，请稍后...
-----------------------------------------------------
E:\PMRF\python\lib\site-packages\realesrgan\utils.py:43: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  loadnet = torch.load(model_path, map_location=torch.device('cpu'))
E:\PMRF\python\lib\site-packages\torchmetrics\utilities\prints.py:43: UserWarning: Metric `InceptionScore` will save all extracted features in buffer. For large datasets this may lead to large memory footprint.
  warnings.warn(*args, **kwargs)  # noqa: B028
Running on local URL:

To create a public link, set `share=True` in `launch()`.
E:\PMRF\python\lib\site-packages\torchvision\models\_utils.py:208: UserWarning: The parameter 'pretrained' is deprecated since 0.13 and may be removed in the future, please use 'weights' instead.
  warnings.warn(
E:\PMRF\python\lib\site-packages\torchvision\models\_utils.py:223: UserWarning: Arguments other than a weight enum or `None` for 'weights' are deprecated since 0.13 and may be removed in the future. The current behavior is equivalent to passing `weights=None`.
  warnings.warn(msg)
E:\PMRF\python\lib\site-packages\facexlib\detection\__init__.py:22: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  load_net = torch.load(model_path, map_location=lambda storage, loc: storage)
E:\PMRF\python\lib\site-packages\facexlib\parsing\__init__.py:20: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See or more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  load_net = torch.load(model_path, map_location=lambda storage, loc: storage)
NATTEN failure: CUDA runtime error: no kernel image is available for execution on the device at: 195
请按任意键继续. . .

-------------------------------------------------------------------------------------------------------------------------------------------------

AI***

同样问题：NATTEN failure: CUDA runtime error: no kernel image is available for execution on the device at: 195

无言***

jiaorong 发表于 2024-10-31 17:59
启动正常，点运行后报错，：“CUDA runtime error: no kernel image is available for execution on the de ...

有个依赖兼容问题，暂时还没找到解决办法，有的人能运行，有的报错。