What chroma upsampling method should be applied to telecined material?

[roozhou]

For 3:2 hard-pulldown YUV 4:2:0 materials, should I treat them as progressive or interlaced when converting to YUY2/RGB?

And what after IVTC? AFAIK ivtc filters do not "resample" interlaced YV12 to progressive YV12. If I encode IVTCed 30t DVD to 24p H264, how can i make sure correct chroma upsampling is applied when playing such H264 videos on a computer?

[tritical]

For hard-telecine material, where pulldown is applied before encoding and the frames are stored as interlaced, you want to match the fields (perform ivtc) and then use progressive chroma upsampling. After matching the fields the chroma positioning will line up correctly for progressive upsampling... see this thread: http://forum.doom9.org/showthread.php?t=79028. Of course, the results wont be as good as if the original material had been encoded progressively and soft-telecine applied using flags. However, it should be better than using interlaced chroma upsampling.

[roozhou]

If the film is first converted to YV12 using progressive chroma downsampling and then hard-telecined and encoded, progressive upsampling after ivtc will be perfect.
But if the film is first separated into fields and then downsampled to interlaced YV12, progressive upsampling will really mess things up.

[tritical]

If the film is first converted to YV12 using progressive chroma downsampling and then hard-telecined and encoded, progressive upsampling after ivtc will be perfect.
But if the film is first separated into fields and then downsampled to interlaced YV12, progressive upsampling will really mess things up.
There is no separating of frames into fields prior to conversion. The interlaced 4:2:2->4:2:0 or yuy2->yv12 conversion operates on full frames. The sampling (interpolated) positions line up so that if you convert a progressive frame using interlaced downsampling you will get the correct result if you use progressive upsampling on it... and that result will be better (in terms of distance to interpolation points used to create new values) than using interlaced upsampling. That is why performing ivtc and then using progressive upsampling is the better than upsampling as interlaced and performing ivtc later. Of course, the difference is not big except for extreme cases. I think all of this is covered in the thread I linked to, but it has to be read pretty thoroughly.

In fact, the chroma positioning in the interlaced down conversion is exactly the same as that of the progressive down conversion (in terms of spatial positions on the frame). The only difference is how you create those values... the progressive down conversion can use whatever samples it wants regardless of whether they come from one field or both fields. While the interlaced conversion is only allowed to use samples from one field or the other.