Still shots don't quite capture the essence of the uncanny valley; how things move. Sturgeon's law will assure that most CG won't get the movement down correctly, so we will at least be weirded out by that. Conversely,
there was a tech demo a little while ago that generated humanlike shapes in real time from multiple cameras:
Mr Kreylos, researcher at University of California, Davis, said in his blog that despite the low quality of the image, it still feels very real.
Again, still images don't quite convey this because movement seems to be a critical factor in how "uncanny" something looks to the human brain.
Escaping the uncanny valley is a matter of presentation, not technology, so I would not expect all CG to suddenly escape the valley. At least, not until computers can predict accurate-enough human movement for us.