| {"node":{"id":"urn:cid:bafkr4ihjtsf3mnk5knzcv5rsql5flz2qm4x2kj7tqw2t3ukgtwyy7kixny","properties":{"nodeType":"data","timestamp":"2024-01-29T16:00:23Z","dataRegistrationJcs":"urn:cid:baga6yaq6eacodu4hepmrm2sogq3kirqr47xi6prup32tytlhkabvsxffqls3s","registeredBy":"did:key:z6MkhQD1A9eMQ8bZNGmBiCVz7kG4mfnApD7WjHKNhkZp7HEY"}},"enrichments":{"asset_hub":{"asset_id":98,"asset_name":"style-transfer-pytorch","owning_project":"","asset_description":"An implementation of neural style transfer in PyTorch, supporting CPUs and Nvidia GPUs. It offers automatic multi-scale stylization for high-quality high-resolution outputs, compatible even up to print resolution. The code supports dual GPU usage for higher maximum resolution. Modifications from the original algorithm include the use of PyTorch pre-trained VGG-19 weights, 'replicate' padding mode in the first layer of VGG-19, scaled results for average/L2 pooling, Wasserstein-2 style loss, an exponential moving average over iterates, warm-starting of the Adam optimizer, non-equal weights for style layers, and progressive scaling of image stylization.","asset_format":"PyTorch","asset_type":"Code","asset_blob_type":"iroh-collection","source_location_url":"","contact_info":"https://twitter.com/RiversHaveWings","license":"MIT","license_link":"https://github.com/crowsonkb/style-transfer-pytorch/blob/master/LICENSE","registered_date":"2024-01-29T16:00:31.86777Z","last_modified_date":"2024-01-29T16:00:31.86777Z"}}} |