Thank you very much for posting your raw sample images, as well as examples of where you ran into trouble with them.
I have now looked at the material and processed the examples myself. The important point is that I do not see a general problem with ColorPerfect when the expected workflows are used.
I am aware that there is a method being propagated by others that uses TouchUp mode to gamma-encode raw linear color-negative captures before using ColorNeg. If people like that approach, they are of course free to use it. I am not here to tell anyone what they may or may not do, and I am not criticizing it simply because it is different.
From my point of view, however, this step is technically unwarranted and a rather bizarre detour. There is no physical basis for it, and especially DC mode will produce faulty output if the input has already been altered in the wrong way. People can do whatever they want if it gives them results they like. But once such a step causes a failure, the reason for that failure needs to be stated clearly.
In a separate follow-up message, I will show more directly what this TouchUp step actually does to the data.
For the first example, I processed your naval scene in ColorNeg DC. Since I did not know which film type it was shot on, I used the built-in characterization for Fuji Superia Reala 100. From memory, this was a film where the three film gammas were very closely aligned, so it was a reasonable starting point.
I expected that, if necessary, I might have to fine-tune the result using the film type / subtype / film gamma calibration system, but in this case that was not necessary. The plain RGB-light TIFF input processed in DC mode already works.
非常感谢你发布原始样张，以及你在处理这些图像时遇到问题的示例。
我现在已经查看了这些材料，并且自己处理了这些例子。重点是：在使用预期工作流程时，我没有看到 ColorPerfect 存在一般性的问题。
我知道现在有人在推广一种方法，即在使用 ColorNeg 之前，先用 TouchUp mode 对原始线性色彩负片翻拍文件进行 gamma 编码。如果有人喜欢这种做法，当然可以自由使用。我并不是要告诉别人什么可以做、什么不可以做，也不是因为它与我的方法不同就批评它。
不过从我的角度来看，这一步在技术上没有必要，而且是一个相当奇怪的绕路。它没有物理基础，尤其是 DC 模式本来就期望输入为线性 gamma；如果输入已经被错误地改变过，那么输出也会相应出错。如果某种做法能得到用户喜欢的结果，大家当然可以按自己的方式来。但一旦这样的步骤导致失败，就必须清楚说明失败的原因。
在后续的另一条消息中，我会更直接地展示这个 TouchUp 步骤实际上对数据做了什么。
第一个例子中，我用 ColorNeg DC 处理了你的海景图像。由于我不知道这张照片使用的是哪种胶片，所以我选择了内置的 Fuji Superia Reala 100 特性化数据。凭我的记忆，这是一种三条胶片 gamma 非常接近的胶片，因此它是一个合理的起点。
我原本预计，如有必要，可能需要使用胶片类型 / 子类型 / 胶片 gamma 校准系统进一步微调结果，但在这个例子中并不需要。普通 RGB 光源 TIFF 输入文件在 DC 模式下已经可以正常处理。
I also processed the decoupled RGB-light TIFF you provided in SC mode, basically treating it as if it were a scan.
我也用 SC 模式处理了你提供的解耦后的 RGB 光源 TIFF，基本上是把它当作扫描文件来处理。
When comparing the two versions, the broader coastline area mainly shows that the two workflows produce somewhat different renderings. Neither result suggests a fundamental processing failure.
对比这两个版本时，较大范围的海岸线区域主要说明这两种工作流程会产生略有不同的呈现效果。两者都不表示存在根本性的处理失败。
The more interesting detail is the Chinese flag on one of the small boats. There your decoupled input seems to work nicely, and I think this part of your method has merit.
For that reason, I would be very interested in a detailed explanation of how this decoupled TIFF was produced.
更有意思的细节是其中一艘小船上的中国国旗。在这里，你的解耦输入看起来效果不错，我认为你这个方法的这一部分是有价值的。
因此，我很有兴趣了解这个解耦后的 TIFF 是如何生成的，希望你能详细说明。
The second example, the street scene with flash illumination, is better exposed than the RGB-light-source data. The RGB-light-source data could easily use about one EV more exposure.
Nonetheless, I found the flash-lit image more difficult to deal with and ended up using the film type characterization system on this one after all.
第二个例子，也就是使用闪光灯照明的街景，比 RGB 光源数据曝光更好。RGB 光源数据其实完全可以再增加大约一档曝光。
尽管如此，我发现这张闪光灯照明的图像反而更难处理，最终还是在这张图上使用了胶片类型特性化系统。
So, for starters, this is just to let you know that I have looked at the files carefully and that the expected workflows do work. The problems you showed are not evidence that ColorPerfect cannot process this material. Rather, they point to the danger of inserting an unwarranted gamma-changing TouchUp step before a workflow that expects linear input.
所以，首先我只是想告诉你，我已经仔细查看了这些文件，而且预期的工作流程确实可以正常工作。你展示的问题并不证明 ColorPerfect 无法处理这些材料。相反，它们指出了另一个风险：在一个本来期望线性输入的工作流程之前，插入一个没有必要、会改变 gamma 的 TouchUp 步骤。