Merge pull request #9 from matajoh/visualizer_refactor

Visualizer refactor

README.md

@@ -107,14 +107,14 @@ dataset = ffn.PixelDataset.create(path_to_image_file, color_space="RGB",
 ## 3D Datasets
 
 This is where the library becomes a bit picky about input data. The
-[`RayDataset`](nerf/ray_dataset.py) supports a set format for data,
+[`ImageDataset`](nerf/image_dataset.py) supports a set format for data,
 and we provide several datasets in this format to play
 with. These datasets are not stored in the repo, but the library will
 automatically download them to the `data` folder when you first requests them
 which you can do like so:
 
 ```python
-dataset = ffn.RayDataset.load("antinous_400.npz", split="train", num_samples=64)
+dataset = ffn.ImageDataset.load("antinous_400.npz", split="train", num_samples=64)
 ```
 
 We recommend you use one of the following (all datasets are provided in 400 and 800 versions):
@@ -231,7 +231,7 @@ It will produce the frames of the following video:
 
 https://user-images.githubusercontent.com/6894931/142744837-382e13b1-d1cf-4305-870a-b64763c73e54.mp4
 
-Another way to visualize what the model has learned is toproduce a
+Another way to visualize what the model has learned is to produce a
 voxelization of the model. This is different from the voxel-based volume
 rendering, in which multiple voxels contribute to a single sample. Rather, it
 is a sparse octree containing voxels at the places the model has determined are

azureml/aml_env.yml

@@ -2,21 +2,21 @@ channels:
     - pytorch
     - nvidia
 dependencies:
-    - python=3.7
-    - cudatoolkit=11.1
-    - pytorch=1.9.0
-    - torchvision=0.10
-    - torchaudio=0.9
-    - pip=20.2
+    - python=3.9
+    - cudatoolkit=11.3
+    - pytorch=1.12.0
+    - torchvision=0.13.0
+    - torchaudio=0.12.0
+    - pip=21.2.4
     - cudnn
     - pip:
         - azureml-defaults
         - azureml-train-core
         - ffmpeg-python
         - matplotlib
-        - numba==0.54.1
-        - numpy==1.20.3
-        - opencv-python-headless>=4.5.3
+        - numba==0.55.2
+        - numpy==1.22.4
+        - opencv-python-headless>=4.5.5
         - progress
         - requests
         - scenepic

fourier_feature_nets/__init__.py

@@ -8,15 +8,30 @@
     MLP,
     PositionalFourierMLP
 )
+from .image_dataset import ImageDataset
 from .nerf_model import NeRF
 from .octree import OcTree
 from .pixel_dataset import PixelDataset
 from .ray_caster import Raycaster
-from .ray_dataset import RayData, RayDataset
+from .ray_dataset import RayDataset
 from .ray_sampler import RaySampler, RaySamples
 from .signal_dataset import SignalDataset
-from .utils import ETABar, interpolate_bilinear, load_model, orbit
+from .utils import (
+    calculate_blend_weights,
+    ETABar,
+    exponential_lr_decay,
+    hemisphere,
+    interpolate_bilinear,
+    load_model,
+    orbit
+)
 from .version import __version__
+from .visualizers import (
+    ActivationVisualizer,
+    ComparisonVisualizer,
+    EvaluationVisualizer,
+    OrbitVideoVisualizer
+)
 from .voxels_model import Voxels
 
 __all__ = ["__version__",
@@ -29,17 +44,25 @@
            "FourierFeatureMLP",
            "PositionalFourierMLP",
            "GaussianFourierMLP",
+           "ImageDataset",
            "Voxels",
+           "calculate_blend_weights",
+           "exponential_lr_decay",
            "interpolate_bilinear",
+           "hemisphere",
            "load_model",
            "orbit",
            "OcTree",
            "PixelDataset",
            "Raycaster",
            "RaySampler",
            "RaySamples",
-           "RayData",
            "RayDataset",
            "Resolution",
            "SignalDataset",
-           "Triangulation"]
+           "Triangulation",
+           "ActivationVisualizer",
+           "ComparisonVisualizer",
+           "EvaluationVisualizer",
+           "OrbitVideoVisualizer",
+           "PatchVisualizer"]

fourier_feature_nets/camera_info.py

@@ -34,6 +34,11 @@ def square(self) -> "Resolution":
         size = min(self.width, self.height)
         return Resolution(size, size)
 
+    @property
+    def ratio(self) -> float:
+        """Aspect ratio."""
+        return self.width / self.height
+
 
 class CameraInfo(NamedTuple("CameraInfo", [("name", str),
                                            ("resolution", Resolution),
@@ -79,6 +84,13 @@ def project(self, positions: np.ndarray) -> np.ndarray:
         points = points[:, :2] / points[:, 2:3]
         return points
 
+    @property
+    def fov_y_degrees(self) -> float:
+        """Y-axis field of view (in degrees) for the camera."""
+        fov_y = (0.5 * self.resolution.width) / self.intrinsics[1, 1]
+        fov_y = 2 * np.arctan(fov_y)
+        return fov_y * 180 / np.pi
+
     @property
     def position(self) -> np.ndarray:
         """Returns the position of the camera in world coordinates."""