simple inference

.gitmodules

@@ -0,0 +1,3 @@
+[submodule "chap9/chap9/synsin"]
+	path = chap9/chap9/synsin
+	url = https://github.com/facebookresearch/synsin

chap9/inference_unseen_image.py

@@ -0,0 +1,67 @@
+import matplotlib.pyplot as plt
+
+import quaternion
+import torch
+import torch.nn as nn
+import torchvision.transforms as transforms
+
+from PIL import Image
+from set_up_model_for_inference import synsin_model
+
+
+def inference(path_to_model, test_image, save_path):
+    model_to_test = synsin_model(path_to_model)
+    # Load the image
+    transform = transforms.Compose([
+        transforms.Resize((256,256)),
+        transforms.ToTensor(),
+        transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
+
+
+    im = Image.open(test_image)
+    im = transform(im)
+
+    # Parameters for the transformation
+    theta = -0.15
+    phi = -0.1
+    tx = 0
+    ty = 0
+    tz = 0.1
+
+    RT = torch.eye(4).unsqueeze(0)
+    # Set up rotation
+    RT[0,0:3,0:3] = torch.Tensor(quaternion.as_rotation_matrix(quaternion.from_rotation_vector([phi, theta, 0])))
+    # Set up translation
+    RT[0,0:3,3] = torch.Tensor([tx, ty, tz])
+
+    batch = {
+        'images' : [im.unsqueeze(0)],
+        'cameras' : [{
+            'K' : torch.eye(4).unsqueeze(0),
+            'Kinv' : torch.eye(4).unsqueeze(0)
+        }]
+    }
+
+    # Generate a new view at the new transformation
+    with torch.no_grad():
+        pred_imgs = model_to_test.model.module.forward_angle(batch, [RT])
+        depth = nn.Sigmoid()(model_to_test.model.module.pts_regressor(batch['images'][0].cuda()))
+
+    fig, axis = plt.subplots(1,3, figsize=(10,20))
+    axis[0].axis('off')
+    axis[1].axis('off')
+    axis[2].axis('off')
+
+    axis[0].imshow(im.permute(1,2,0) * 0.5 + 0.5)
+    axis[0].set_title('Input Image')
+    axis[1].imshow(pred_imgs[0].squeeze().cpu().permute(1,2,0).numpy() * 0.5 + 0.5)
+    axis[1].set_title('Generated Image')
+    axis[2].imshow(depth.squeeze().cpu().clamp(max=0.04))
+    axis[2].set_title('Predicted Depth')
+
+    plt.savefig(save_path)
+
+if __name__ == '__main__':
+    inference(path_to_model= './synsin/modelcheckpoints/realestate/zbufferpts.pth',
+              test_image='appartement.JPG',
+              save_path='output.png')

chap9/set_up_model_for_inference.py

@@ -0,0 +1,38 @@
+import torch
+import torch.nn as nn
+
+import sys
+sys.path.insert(0, './synsin')
+
+import os
+os.environ['DEBUG'] = '0'
+from synsin.models.networks.sync_batchnorm import convert_model
+from synsin.models.base_model import BaseModel
+from synsin.options.options import get_model
+
+# Set up the models
+def synsin_model(model_path):
+
+    torch.backends.cudnn.enabled = True
+
+    opts = torch.load(model_path)['opts']
+    opts.render_ids = [1]
+
+    model = get_model(opts)
+
+    torch_devices = [int(gpu_id.strip()) for gpu_id in opts.gpu_ids.split(",")]
+
+    if 'sync' in opts.norm_G:
+        model = convert_model(model)
+        model = nn.DataParallel(model, torch_devices[0:1]).cuda()
+    else:
+        model = nn.DataParallel(model, torch_devices[0:1]).cuda()
+
+    #  Load the original model to be tested
+    model_to_test = BaseModel(model, opts)
+    model_to_test.load_state_dict(torch.load(model_path)['state_dict'])
+    model_to_test.eval()
+
+    print("Loaded model")
+
+    return model_to_test