The growing usage of identity documents recognition in day-to-day life increases the need in public datasets with identity documents in various languages. Still, public identity documents datasets mainly cover Latin script-based languages. At the same time, newly emerging countries demonstrate wide adoption of technology in combination with population growth. As a result, competitive identity documents recognition system should include support documents of these countries. In this paper, we present MIDV-UP, where U stands for Urdu and P for Persian, a dataset containing identity documents in these languages. In total, MIDV-UP contains 1000 unique mock identity documents of 4 types, and 9000 fully annotated images of these documents, acquired with different equipment in various conditions. We also provide baselines for the several important tasks in typical identity documents recognition pipeline: document location and identification, per-field segmentation, and fields recognition. The results of the experiments show that the accuracy of the published approaches for the presented identity documents is low, especially in text line recognition task, and that MIDV-UP presents challenges for future research.