How do you measure RX IMR? Do we input two tones and measure the wanted and image signal in the baseband?

you need to input just one signal at the image frequency at the RF input, which causes an in-band response in your receiver. The ratio of this image stimulated in-band response versus a wanted input signal response gives you the image rejection.

The input RF image frequency depends on your architecture... for a super-het, it is offset from the LO by the IF (opposite side to the wanted frequency).