The present application discloses a mask pattern determination method, a mask pattern determination apparatus, a medium, and a product, which are applied to the technical field of semiconductor. In this method, etching bias is determined by a first etch pattern and a first photolithography pattern, and the first etch pattern is obtained by merging the first sub-image and the second sub-image. The first sub-image is generated by a physical effect simulation unit based on the first photolithography pattern, and the second sub-image is generated by a neural network unit based on the first photolithography pattern. The neural network unit can effectively improve a data fitting ability of an etching model, thereby improving an accuracy of the etching model to obtain an accurate first etching bias. Meanwhile, the physical effect simulation unit of the etching model simulates an actual physical effect, which can prevent the etching model from over-fitting.