Cryo-electron tomography (Cryo-ET) enables high-resolution three-dimensional imaging of macromolecules in their native cellular environments. However, the low signal-to-noise ratio, complex cellular environment, and significant differences in the size of macromolecules make it extremely challenging to efficiently and accurately locate and classify these particles. To address these issues, we propose a weakly supervised framework, termed Full-Size Picker (FSPicker), for automated particle picking in 3D tomograms. FSPicker features an innovative network design that combines global contextual perception with local structural refinement, enabling it to effectively suppress noise while capturing fine particle details. In addition, FSPicker adopts a weakly supervised training strategy, which requires only a small number of simplified labels, thus reducing reliance on high-precision manual annotations. Comprehensive evaluations of both simulated and real tomograms demonstrate that FSPicker outperforms existing mainstream methods, particularly in particle detection and localization of small particles under low-SNR conditions, highlighting its superior performance in complex cellular environments. Our code is available at https://github.com/Tianwenhao116/FSPicker .