Deep Image Analogy | PatchMatch | NNF | VGG19 | Image Generation | Visual Attribute Transfer| python
Deep Image Analogy: Visual Attribute Transfer at different Layers of a pretrained VGG19 with the PatchMatch Algorithm with python / pytorch
Given input: Acouple of images A, B' (the left and the right images)
To be output: A couple of images A' and B (two images in the middle) to be generated so that the following (analogy) semantic similarity is maintained: A:A' :: B:B'
Patchmatch algorithm is used for reconstruction, using the ANN (approx nearest neighbor) field computed from the features extracted at different layers for the input images, starting from finer to coarser layers in the pretrained deep neural net VGG19
224 × 224 low resolution images were used as input (as expected by VGG19)
Ref: https://arxiv.org/pdf/1705.01088
#imageprocessing #imageprocessingpython #python #computervision #algorithm #deeplearning #pytorch #generativeart