Terahertz waves, positioned between microwaves and infrared in the electromagnetic spectrum, stand out with their potent penetration, minimal energy, and consistent absorption profiles for specific substances. Their versatility in applications like non-destructive evaluation, human security scans, and biological diagnostics has propelled them to the forefront of scientific inquiry. Yet, the current state of terahertz equipment, both for generation and acquisition, imposes limitations such as compromised resolution, blurring from diffraction, and clarity degradation due to texture overlaps. As a result, many extant multi-view 3D reconstruction algorithms struggle to yield quality results with terahertz imagery. To circumvent these issues, and specifically the scarcity of terahertz datasets and the conflation of textures, we combined X-ray images from the DTU dataset with our gathered terahertz projections. Observing the inconsistent texture projections in multi-view terahertz images based on angle shifts, we conceptualized the DETransMVSnet -a cutting-edge multi-view 3D reconstruction rooted in the Multi-Scale Deep Equilibrium Layer (MDEQ) paradigm. By tapping into the equilibrium layers of homography-projected feature maps, we can extract masks distinguishing the scene's layers. The Intra-Attention and Mask-Attention Blocks further refine feature selection, retaining pertinent terahertz details while sidelining disruptive background elements. As a testament to its efficacy, DETransMVSnet matches conventional algorithms on the DTU dataset but notably outperforms in terahertz datasets, aptly reconstructing images where predecessors faltered.