sashatr/textures

Using a pretrained convolutional network to get texture

Textures

---

The pretrained convolutional network (VGG 19) is used to get texture.

For getting the Style Features are used layers from the list (conv1_1, conv2_1, conv3_1, conv4_1, conv5_1) based on original paper.

---

Below is an example of the results for full images and patches

Origin patches:

---

Using two first layers for patches

---

Origin images:

---

Using all layers

Merger tensor by maximum values Merger tensor by mean

---

Using first layer

Merger tensor by maximum values Merger tensor by mean

---

Using two first layers

Merger tensor by maximum values Merger tensor by mean

---

TODO:

• add padding to rectangular images
• etс