You can try to process the geometrically corrected imagery, but it is highly likely to get strange results with erroneous outputs.
The overall reason for the erroneous outputs is that distortion correction is not a standardized procedure that is uniformly employed across platforms. Often times the exact mathematics of this procedure is not publicly available and can vary from company to company. Thus, these corrections are not taken into account in the modeling of the camera parameters. Pix4Dmapper models the distortion of a camera by default and expects that distortion introduced by a lens to be present. When the distortion is removed prior to processing in Pix4Dmapper, then there are essentially two steps of distortion correction being applied in series, which is not ideal.
The reason that this may fail in Pix4Dmapper is that we place a very high value on accuracy of reconstruction and have a high threshold for what gets recreated with the resulting model. The inaccuracies introduced by pre-processing the images can skew the results to a level of inaccuracy that Pix4Dmapper is not going to reconstruct. Other applications that focus more on just generating something visually pleasant with the price of less accuracy may be able to complete processing this type of scenario.
As of now, processing undistorted images is not a supported workflow that we recommend for Pix4Dmapper. I would just recommend that you try it yourself to see the difference in the results.
As the correction of distorted images prior to importing them into Pix4Dmapper is out of my technical jurisdiction, I will let some other savvy users explain more about this process.
Hopefully this helps.