From 1e8072ce3dda94be2f7961fc5f243dc7a3453910 Mon Sep 17 00:00:00 2001
From: Martin Rauscher <hades32@gmail.com>
Date: Sat, 23 Jun 2018 19:24:17 +0200
Subject: [PATCH 1/2] Created using Colaboratory

---
 demo.ipynb | 41 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 41 insertions(+)
 create mode 100644 demo.ipynb

diff --git a/demo.ipynb b/demo.ipynb
new file mode 100644
index 0000000..479460d
--- /dev/null
+++ b/demo.ipynb
@@ -0,0 +1,41 @@
+{
+  "nbformat": 4,
+  "nbformat_minor": 0,
+  "metadata": {
+    "colab": {
+      "name": "Untitled0.ipynb",
+      "version": "0.3.2",
+      "provenance": [],
+      "include_colab_link": true
+    },
+    "kernelspec": {
+      "name": "python3",
+      "display_name": "Python 3"
+    }
+  },
+  "cells": [
+    {
+      "cell_type": "markdown",
+      "metadata": {
+        "id": "view-in-github",
+        "colab_type": "text"
+      },
+      "source": [
+        "[View in Colaboratory](https://colab.research.google.com/github/Hades32/MegaDepth/blob/master/demo.ipynb)"
+      ]
+    },
+    {
+      "metadata": {
+        "id": "ViJt7oifpX3d",
+        "colab_type": "code",
+        "colab": {}
+      },
+      "cell_type": "code",
+      "source": [
+        ""
+      ],
+      "execution_count": 0,
+      "outputs": []
+    }
+  ]
+}
\ No newline at end of file

From d4c8cb21398bc87639953f570d90f66d9ff564c9 Mon Sep 17 00:00:00 2001
From: Martin Rauscher <hades32@gmail.com>
Date: Sun, 24 Jun 2018 00:13:48 +0200
Subject: [PATCH 2/2] set PyTorch in non-training mode. Faster, less RAM usage

---
 demo.py | 47 ++++++++++++++++++++++++-----------------------
 1 file changed, 24 insertions(+), 23 deletions(-)

diff --git a/demo.py b/demo.py
index 2447470..00dd182 100644
--- a/demo.py
+++ b/demo.py
@@ -22,29 +22,30 @@ def test_simple(model):
     total_loss =0 
     toal_count = 0
     print("============================= TEST ============================")
-    model.switch_to_eval()
-
-    img = np.float32(io.imread(img_path))/255.0
-    img = resize(img, (input_height, input_width), order = 1)
-    input_img =  torch.from_numpy( np.transpose(img, (2,0,1)) ).contiguous().float()
-    input_img = input_img.unsqueeze(0)
-
-    input_images = Variable(input_img.cuda() )
-    pred_log_depth = model.netG.forward(input_images) 
-    pred_log_depth = torch.squeeze(pred_log_depth)
-
-    pred_depth = torch.exp(pred_log_depth)
-
-    # visualize prediction using inverse depth, so that we don't need sky segmentation (if you want to use RGB map for visualization, \
-    # you have to run semantic segmentation to mask the sky first since the depth of sky is random from CNN)
-    pred_inv_depth = 1/pred_depth
-    pred_inv_depth = pred_inv_depth.data.cpu().numpy()
-    # you might also use percentile for better visualization
-    pred_inv_depth = pred_inv_depth/np.amax(pred_inv_depth)
-
-    io.imsave('demo.png', pred_inv_depth)
-    # print(pred_inv_depth.shape)
-    sys.exit()
+    with torch.no_grad():
+        model.switch_to_eval()
+
+        img = np.float32(io.imread(img_path))/255.0
+        img = resize(img, (input_height, input_width), order = 1)
+        input_img =  torch.from_numpy( np.transpose(img, (2,0,1)) ).contiguous().float()
+        input_img = input_img.unsqueeze(0)
+
+        input_images = Variable(input_img.cuda() )
+        pred_log_depth = model.netG.forward(input_images) 
+        pred_log_depth = torch.squeeze(pred_log_depth)
+
+        pred_depth = torch.exp(pred_log_depth)
+
+        # visualize prediction using inverse depth, so that we don't need sky segmentation (if you want to use RGB map for visualization, \
+        # you have to run semantic segmentation to mask the sky first since the depth of sky is random from CNN)
+        pred_inv_depth = 1/pred_depth
+        pred_inv_depth = pred_inv_depth.data.cpu().numpy()
+        # you might also use percentile for better visualization
+        pred_inv_depth = pred_inv_depth/np.amax(pred_inv_depth)
+
+        io.imsave('demo.png', pred_inv_depth)
+        # print(pred_inv_depth.shape)
+        sys.exit()