Additive manufacturing (AM) is a new paradigm in designdriven build of customized products. Nonetheless, mass customization and low volume production make the AM qualityassurance extremely challenging. Advanced imaging providesan unprecedented opportunity to increase information visibility,cope with the product complexity, and enable on-the-fly qualitycontrol in AM. However, in-situ images of a customized AM buildshow a high level of layer-to-layer geometry variation, whichhampers the use of powerful image-based learning methods suchas deep neural networks (DNNs) for flaw detection. Few, if any,previous works investigated how to tackle the impact of AM customization on image-guided process monitoring and control. Theproposed research is aimed at filling this gap by developing anovel real-time and multi-scale process monitoring methodology for quality control of customized AM builds. Specifically, weleverage the computer-aided design (CAD) file to perform shapeto-image registration and delineate the regions of interests in layerwise images. Next, a hierarchical dyadic partitioning methodology is developed to split layer-to-layer regions of interest intosubregions with the same number of pixels to provide freeform geometry analysis. Then, we propose a semiparametric modelto characterize the complex spatial patterns in each customizedsubregion and boost the computational speed. Finally, a DNNmodel is designed to learn and detect fine-grained informationof flaws. Experimental results show that the proposed processmonitoring and control methodology detects flaws in each layerwith an accuracy of 92:50±1:03%. This provides an opportunityto reduce inter-layer variation in AM prior to completion of thebuild. The proposed methodology can also be generally applicable in a variety of engineering and medical domains that entailimage-based process monitoring and control with customized designs.