| # Vladimir Petrovich Ermolaev: Photography, Memory, and a Learned Print Aesthetic | |
| Vladimir Petrovich Ermolaev (28 July 1892 – 1 April 1982) was an ethnographer, journalist, museum worker, and one of the most important photographic chroniclers of twentieth-century Tuva. He first visited Tuva as a photographer in 1913 and later settled there permanently. In 1929 he became the first director of the Tuvan State Museum, helping to establish its collections and exhibitions. His photographs documented everyday life, public figures, political change, industry, transport, culture, wartime participation, and the transformation of the Tuvan People’s Republic. | |
| Collections of Ermolaev’s photographic albums are preserved by the National Archives of the Republic of Tuva. They remain an important visual record for researchers and the public. The archive’s public collection catalogue can be explored at [fond.gosarhivrt.ru](https://fond.gosarhivrt.ru/), and the National Archives has published a biographical article marking the 130th anniversary of his birth. | |
| The model named **Ermolaev** is a tribute to this photographic legacy. It learns an aesthetic domain rather than reconstructing a specific historical camera, negative, event, or original photograph. A crucial material fact shaped the dataset: the historical domain was prepared from **paper photographic copies**, not original negatives or glass plates. Paper fibres, printing processes, copying, tonal compression, surface texture, fading, and accumulated physical history are therefore visible in the learned appearance. The result should be understood as a learned historical-print aesthetic with pronounced texture, not as a neutral simulation of an original photographic negative. | |
| The model was created using Contrastive Unpaired Translation (CUT). Modern photographs and historical print images were supplied as unpaired domains, allowing the generator to learn a mapping toward the tonal and textural character of the collection while attempting to preserve scene structure. Like all generative image translation, it may alter fine details, faces, text, and culturally significant attributes. Its output is an interpretation and must not be presented as authentic archival evidence. | |
| Dataset author and model developer: **Valeriy Aldyn-oolovich Irgit**. | |
| The training images are not redistributed in this model repository. Researchers wishing to study the historical collection should use the National Archives catalogue and follow the archive’s access and reuse conditions. | |