add biggan test

justheuristic · web-flow · commit 0717a01857c5 · 2020-03-24T19:31:20.000+03:00
diff --git a/seminar06-style-transfer/README.md b/seminar06-style-transfer/README.md
@@ -1 +1,32 @@
 [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/yandexdataschool/Practical_DL/blob/spring20/seminar06-style-transfer/style_transfer_pytorch.ipynb)
+
+
+
+`to be used in seminar`
+
+```(python)
+import multiprocessing as mp
+import multiprocessing.shared_memory
+import numpy as np
+import torch
+from pytorch_pretrained_biggan import BigGAN
+import matplotlib.pyplot as plt
+%matplotlib inline
+model = BigGAN.from_pretrained('biggan-deep-128').cuda().eval()
+model.cuda();
+model();
+with torch.no_grad():
+
+    z_torch = torch.randn(1, 128)
+    c_torch = torch.randint(0, 1000, size=[1])
+
+    gen_images = model.forward(
+        z_torch.clamp_(-5, 5).cuda(), 
+        torch.nn.functional.one_hot(c_torch.clamp_(0, 999).cuda(), num_classes=1000).to(torch.float32),
+        truncation=0.4)
+
+    gen_images = torch.clamp((gen_images + 1) / 2, 0, 1)
+    image = gen_images.data.cpu().numpy()[0].transpose(1, 2, 0)
+
+    plt.imshow(image)
+```
