FSR does not convince
The loser of the three techniques is quickly found: AMD's FSR is clearly inferior to both DLSS and TSR. And that's not surprising, because as a spatial upscaling, FSR, unlike the other two methods, cannot access the information from previous frames. One of the consequences of this is that there is no temporal reconstruction. Fine lines, some of which are no longer displayed correctly with the native resolution, lose even more structure with FSR.
On top of that, FSR gets very muddy in low resolutions. Since FSR sharpens properly on "Ultra Quality" and FSR on "Quality" in Ultra HD, this is not noticeable in either mode. However, all other or a lower target resolution clearly lose sharpness and thus details. Since the game's own anti-aliasing works well, the temporal stability in Ultra HD with FSR UQ and FSR Q is not a big problem, but the graphics then flicker visibly with more aggressive settings.
TSR has problems but makes a good impression
TSR is consistently preferable to FSR in terms of image quality. Always. In every resolution and with every setting. The temporal peculiarities bring many advantages, there are only a few disadvantages in Ghostwire: Tokyo. For example, TSR can restore fine lines and is significantly better in this regard, even in Ultra HD and in the performance setting and thus a rendering resolution of Full HD, than native Ultra HD with four times the number of pixels.
In addition, there is hardly any loss of image sharpness with TSR, even with TSR set to "Performance" in Ultra HD, the image is just as sharp as with the native Ultra HD resolution. And the image stability is also very good, only with TSR on "Performance" does it get worse in Ultra HD and there is slight flickering, which is still better than with FSR on "Ultra Quality".
However, there are two disadvantages with TSR: The upsampling cannot reconstruct the raytracing reflections well, there just as many details are lost compared to the native resolution as with FSR (and DLSS). And some objects show ghosting in motion, which is good to see in places. With TSR on "Quality" this is still within limits in Ultra HD, but with TSR on "Performance" it is clearly noticeable.
DLSS is the winner, but only just
TSR and Nvidia's DLSS then differ optically in more detail, both technologies have their advantages and disadvantages. DLSS offers the same advantages as TSR in terms of temporal stability, reconstruction and image sharpness at low resolutions and is slightly better in terms of reconstruction at aggressive settings such as "Performance" in Ultra HD or a lower target resolution. However, the differences are small and non-existent in terms of image sharpness.
On the other hand, TSR has small advantages when it comes to temporal stability. DLSS flickers slightly more on aggressive settings, but this only applies to individual objects in the game. With a view to ghosting, DLSS has the edge, but is also not completely free of the problem. With TSR, more objects are simply affected by the problem than with DLSS. Finally, it's worth noting that from DLSS on "Performance" in Ultra HD or a target resolution of WQHD and below, some ray-traced reflections flicker strangely, which is not the case with TSR and FSR.