[Objective] Large-sized rockfill forms the primary load-bearing skeleton in rock-filled concrete (RFC), and its morphological characteristics strongly influence the mechanical performance of dam structures. This study develops a practical 3D reconstruction method for large-sized rockfill under field engineering conditions, aiming to reveal the similarities and differences in shape parameters and provide a reliable technical approach and foundational data for the selection, evaluation, and performance optimization of rockfill materials in RFC engineering. [Methods] A multi-view 3D reconstruction technique based on structure from motion (SFM) was employed to acquire and reconstruct the morphological characteristics of large-sized rockfill. The segment anything model (SAM) was employed to isolate rockfill from the background in the images. For morphological analysis, the reconstructed models were imported into Blender software. Based on established particle shape characterization systems, parameters including equivalent particle size, elongation, flatness, roundness, true sphericity, convexity, and maximum projected sphericity were quantified. Following the Zingg scheme, particles were categorized into spherical, platy, bladed, and rod-like forms. Statistical analysis involved cumulative distribution function (CDF) curves to characterize overall parameter trends and box plots to compare parameter distributions across different particle size groups. Additionally, morphological data from the YF project were compared with results from the DZ project reported in existing literature to examine the similarities and differences between projects with different lithologies and particle size distributions. [Results] Analysis of 72 reconstructed specimens from the YF project revealed that the rockfill exhibited a generally blocky and well-rounded macroscopic morphology. Statistical shape classification showed that 45% of the YF rockfill was spherical, 23% was platy, 21% was rod-like, and 11% was bladed. Comparative analysis among different particle size groups within the YF project identified specific surface area as the parameter most sensitive to particle size, displaying a nearly linear monotonic decreasing trend with increasing particle size. Elongation showed a monotonic increase with particle size, while convexity showed a monotonic decrease. The overall distribution patterns of YF (granite) and DZ (limestone) particles in the shape classification were similar between the two projects. The relationships between key parameters, including the power-function correlations between the length-to-width ratio and roundness and between volume and surface area, were also consistent. However, the YF project exhibited a lower proportion of spherical particles and higher proportions of platy, bladed, and rod-like particles, indicating slightly greater shape irregularity and diversity. [Conclusion] (1) The proposed 3D reconstruction methodology for large-sized rockfill under field engineering conditions has been validated effective and practical. (2) The morphological analysis of 72 rockfill specimens from the YF project confirms that the field rockfill is predominantly well-rounded and blocky. While parameters such as roundness and true sphericity show high values with limited sensitivity to particle size, the specific surface area demonstrates clear systematic variation across different size groups. (3) Despite differences in lithology, the rockfill from the YF and DZ projects shows strong consistency in fundamental macroscopic morphological characteristics and the functional relationships between key shape parameters. (4) The shape composition of the YF project rockfill differs from that of the DZ project, with a lower percentage of spherical particles and higher percentages of non-spherical (platy, bladed, and rod-like) particles.