Abstract: A method for transfer learning is provided. A first set of image data at a first resolution is received. Metadata, comprising object data, associated with the first set of image data is received. A second set of images having a second resolution is generated by downsampling the first set of image data. A third set of image data having a third resolution is generated by upscaling the second set of image data. The third set of image data is associated with the metadata associated to the first set of image data. A computer implemented model is trained to detect objects using the combination of the third set of image data and the associated metadata. A fourth set of image data at the third resolution is received. The trained computer implemented model is refined using the fourth set of image data.