Based on the assumption that the time evolution of a sample observed by computed tomography requires many less parameters than the definition of the microstructure itself, it is proposed to reconstruct these changes based on the initial state (using computed tomography) and very few radiographs acquired at fixed intervals of time. This Letter presents a proof of concept that for a fatigue cracked sample its kinematics can be tracked from no more than two radiographs in situations where a complete 3D view would require several hundreds of radiographs. This 2 order of magnitude gain opens the way to a "computed" 4D tomography, which complements the recent progress achieved in fast or ultrafast computed tomography, which is based on beam brightness, detector sensitivity, and signal acquisition technologies.