Return R2N2 R,T,K

Summary: Return rotation, translation and intrinsic matrices necessary to reproduce R2N2's own renderings.

Reviewed By: nikhilaravi

Differential Revision: D22462520

fbshipit-source-id: 46a3859743ebc43c7a24f75827d2be3adf3f486b

pytorch3d/datasets/r2n2/r2n2.py

@@ -10,7 +10,9 @@ import numpy as np
 import torch
 from PIL import Image
 from pytorch3d.datasets.shapenet_base import ShapeNetBase
+from pytorch3d.datasets.utils import compute_extrinsic_matrix
 from pytorch3d.io import load_obj
+from pytorch3d.renderer import HardPhongShader
 from pytorch3d.renderer.cameras import CamerasBase
 from pytorch3d.transforms import Transform3d
 from tabulate import tabulate
@@ -168,6 +170,9 @@ class R2N2(ShapeNetBase):
             - label (str): synset label.
             - images: FloatTensor of shape (V, H, W, C), where V is number of views
                 returned. Returns a batch of the renderings of the models from the R2N2 dataset.
+            - R: Rotation matrix of shape (V, 3, 3), where V is number of views returned.
+            - T: Translation matrix of shape (V, 3), where V is number of views returned.
+            - K: Intrinsic matrix of shape (V, 4, 4), where V is number of views returned.
         """
         if isinstance(model_idx, tuple):
             model_idx, view_idxs = model_idx
@@ -213,7 +218,11 @@ class R2N2(ShapeNetBase):
                 "rendering",
             )
 
-            images = []
+            # Read metadata file to obtain params for calibration matrices.
+            with open(path.join(rendering_path, "rendering_metadata.txt"), "r") as f:
+                metadata_lines = f.readlines()
+
+            images, Rs, Ts = [], [], []
             for i in model_views:
                 # Read image.
                 image_path = path.join(rendering_path, "%02d.png" % i)
@@ -221,10 +230,125 @@ class R2N2(ShapeNetBase):
                 image = torch.from_numpy(np.array(raw_img) / 255.0)[..., :3]
                 images.append(image.to(dtype=torch.float32))
 
+                # Get camera calibration.
+                azim, elev, yaw, dist_ratio, fov = [
+                    float(v) for v in metadata_lines[i].strip().split(" ")
+                ]
+                R, T = self._compute_camera_calibration(azim, elev, dist_ratio)
+                Rs.append(R)
+                Ts.append(T)
+
+            # Intrinsic matrix extracted from the Blender with slight modification to work with
+            # PyTorch3D world space. Taken from meshrcnn codebase:
+            # https://github.com/facebookresearch/meshrcnn/blob/master/shapenet/utils/coords.py
+            K = torch.tensor(
+                [
+                    [2.1875, 0.0, 0.0, 0.0],
+                    [0.0, 2.1875, 0.0, 0.0],
+                    [0.0, 0.0, -1.002002, -0.2002002],
+                    [0.0, 0.0, 1.0, 0.0],
+                ]
+            )
             model["images"] = torch.stack(images)
+            model["R"] = torch.stack(Rs)
+            model["T"] = torch.stack(Ts)
+            model["K"] = K.expand(len(model_views), 4, 4)
 
         return model
 
+    def _compute_camera_calibration(self, azim: float, elev: float, dist_ratio: float):
+        """
+        Helper function for calculating rotation and translation matrices from azimuth
+        angle, elevation and distance ratio.
+
+        Args:
+            azim: Rotation about the z-axis, in degrees.
+            elev: Rotation above the xy-plane, in degrees.
+            dist_ratio: Ratio of distance from the origin to the maximum camera distance.
+
+        Returns:
+            - R: Rotation matrix of shape (3, 3).
+            - T: Translation matrix of shape (3).
+        """
+        # Retrive R,T,K of the selected view(s) by reading the metadata.
+        MAX_CAMERA_DISTANCE = 1.75  # Constant from R2N2.
+        dist = dist_ratio * MAX_CAMERA_DISTANCE
+        RT = compute_extrinsic_matrix(azim, elev, dist)
+
+        # Transform the mesh vertices from shapenet world to pytorch3d world.
+        shapenet_to_pytorch3d = torch.tensor(
+            [
+                [-1.0, 0.0, 0.0, 0.0],
+                [0.0, 1.0, 0.0, 0.0],
+                [0.0, 0.0, -1.0, 0.0],
+                [0.0, 0.0, 0.0, 1.0],
+            ],
+            dtype=torch.float32,
+        )
+        RT = compute_extrinsic_matrix(azim, elev, dist)  # (4, 4)
+        RT = torch.transpose(RT, 0, 1).mm(shapenet_to_pytorch3d)  # (4, 4)
+
+        # Extract rotation and translation matrices from RT.
+        R = RT[:3, :3]
+        T = RT[3, :3]
+        return R, T
+
+    def render(
+        self,
+        model_ids: Optional[List[str]] = None,
+        categories: Optional[List[str]] = None,
+        sample_nums: Optional[List[int]] = None,
+        idxs: Optional[List[int]] = None,
+        view_idxs: Optional[List[int]] = None,
+        shader_type=HardPhongShader,
+        device="cpu",
+        **kwargs
+    ) -> torch.Tensor:
+        """
+        Render models with BlenderCamera by default to achieve the same orientations as the
+        R2N2 renderings. Also accepts other types of cameras and any of the args that the
+        render function in the ShapeNetBase class accepts.
+
+        Args:
+            view_idxs: each model will be rendered with the orientation(s) of the specified
+                views. Only render by view_idxs if no camera or args for BlenderCamera is
+                supplied.
+            Accepts any of the args of the render function in ShapnetBase:
+            model_ids: List[str] of model_ids of models intended to be rendered.
+            categories: List[str] of categories intended to be rendered. categories
+                and sample_nums must be specified at the same time. categories can be given
+                in the form of synset offsets or labels, or a combination of both.
+            sample_nums: List[int] of number of models to be randomly sampled from
+                each category. Could also contain one single integer, in which case it
+                will be broadcasted for every category.
+            idxs: List[int] of indices of models to be rendered in the dataset.
+            shader_type: Shader to use for rendering. Examples include HardPhongShader
+            (default), SoftPhongShader etc or any other type of valid Shader class.
+            device: torch.device on which the tensors should be located.
+            **kwargs: Accepts any of the kwargs that the renderer supports and any of the
+                args that BlenderCamera supports.
+
+        Returns:
+            Batch of rendered images of shape (N, H, W, 3).
+        """
+        idxs = self._handle_render_inputs(model_ids, categories, sample_nums, idxs)
+        r = torch.cat([self[idxs[i], view_idxs]["R"] for i in range(len(idxs))])
+        t = torch.cat([self[idxs[i], view_idxs]["T"] for i in range(len(idxs))])
+        k = torch.cat([self[idxs[i], view_idxs]["K"] for i in range(len(idxs))])
+        # Initialize default camera using R, T, K from kwargs or R, T, K of the specified views.
+        blend_cameras = BlenderCamera(
+            R=kwargs.get("R", r),
+            T=kwargs.get("T", t),
+            K=kwargs.get("K", k),
+            device=device,
+        )
+        cameras = kwargs.get("cameras", blend_cameras).to(device)
+        kwargs.pop("cameras", None)
+        # pass down all the same inputs
+        return super().render(
+            idxs=idxs, shader_type=shader_type, device=device, cameras=cameras, **kwargs
+        )
+
 
 class BlenderCamera(CamerasBase):
     """

pytorch3d/datasets/shapenet_base.py

@@ -111,12 +111,27 @@ class ShapeNetBase(torch.utils.data.Dataset):
         Returns:
             Batch of rendered images of shape (N, H, W, 3).
         """
-        paths = self._handle_render_inputs(model_ids, categories, sample_nums, idxs)
+        idxs = self._handle_render_inputs(model_ids, categories, sample_nums, idxs)
+        paths = [
+            path.join(
+                self.shapenet_dir,
+                self.synset_ids[idx],
+                self.model_ids[idx],
+                self.model_dir,
+            )
+            for idx in idxs
+        ]
         meshes = load_objs_as_meshes(paths, device=device, load_textures=False)
         meshes.textures = TexturesVertex(
             verts_features=torch.ones_like(meshes.verts_padded(), device=device)
         )
         cameras = kwargs.get("cameras", OpenGLPerspectiveCameras()).to(device)
+        if len(cameras) != 1 and len(cameras) % len(meshes) != 0:
+            raise ValueError("Mismatch between batch dims of cameras and meshes.")
+        if len(cameras) > 1:
+            # When rendering R2N2 models, if more than one views are provided, broadcast
+            # the meshes so that each mesh can be rendered for each of the views.
+            meshes = meshes.extend(len(cameras) // len(meshes))
         renderer = MeshRenderer(
             rasterizer=MeshRasterizer(
                 cameras=cameras,
@@ -136,7 +151,7 @@ class ShapeNetBase(torch.utils.data.Dataset):
         categories: Optional[List[str]] = None,
         sample_nums: Optional[List[int]] = None,
         idxs: Optional[List[int]] = None,
-    ) -> List[str]:
+    ) -> List[int]:
         """
         Helper function for converting user provided model_ids, categories and sample_nums
         to indices of models in the loaded dataset. If model idxs are provided, we check if
@@ -206,15 +221,7 @@ class ShapeNetBase(torch.utils.data.Dataset):
                 )
                 warnings.warn(msg)
             idxs = self._sample_idxs_from_category(sample_nums[0])
-        return [
-            path.join(
-                self.shapenet_dir,
-                self.synset_ids[idx],
-                self.model_ids[idx],
-                self.model_dir,
-            )
-            for idx in idxs
-        ]
+        return idxs
 
     def _sample_idxs_from_category(
         self, sample_num: int = 1, category: Optional[str] = None

pytorch3d/datasets/utils.py

 # Copyright (c) Facebook, Inc. and its affiliates. All rights reserved.
-
+import math
 from typing import Dict, List
 
 import torch
@@ -34,10 +34,77 @@ def collate_batched_meshes(batch: List[Dict]):
             verts=collated_dict["verts"], faces=collated_dict["faces"]
         )
 
-    # If collate_batched_meshes receives R2N2 items, stack the batches of
-    # views of each model into a new batch of shape (N, V, H, W, 3) where
-    # V is the number of views.
+    # If collate_batched_meshes receives R2N2 items with images and that
+    # all models have the same number of views V, stack the batches of
+    # views of each model into a new batch of shape (N, V, H, W, 3).
+    # Otherwise leave it as a list.
     if "images" in collated_dict:
-        collated_dict["images"] = torch.stack(collated_dict["images"])
+        try:
+            collated_dict["images"] = torch.stack(collated_dict["images"])
+        except RuntimeError:
+            print(
+                "Models don't have the same number of views. Now returning "
+                "lists of images instead of batches."
+            )
+
+    # If collate_batched_meshes receives R2N2 items with camera calibration
+    # matrices and that all models have the same number of views V, stack each
+    # type of matrices into a new batch of shape (N, V, ...).
+    # Otherwise leave them as lists.
+    if all(x in collated_dict for x in ["R", "T", "K"]):
+        try:
+            collated_dict["R"] = torch.stack(collated_dict["R"])  # (N, V, 3, 3)
+            collated_dict["T"] = torch.stack(collated_dict["T"])  # (N, V, 3)
+            collated_dict["K"] = torch.stack(collated_dict["K"])  # (N, V, 4, 4)
+        except RuntimeError:
+            print(
+                "Models don't have the same number of views. Now returning "
+                "lists of calibration matrices instead of batches."
+            )
 
     return collated_dict
+
+
+def compute_extrinsic_matrix(azimuth, elevation, distance):
+    """
+    Copied from meshrcnn codebase:
+    https://github.com/facebookresearch/meshrcnn/blob/master/shapenet/utils/coords.py#L96
+
+    Compute 4x4 extrinsic matrix that converts from homogenous world coordinates
+    to homogenous camera coordinates. We assume that the camera is looking at the
+    origin.
+    Used in R2N2 Dataset when computing calibration matrices.
+
+    Args:
+        azimuth: Rotation about the z-axis, in degrees.
+        elevation: Rotation above the xy-plane, in degrees.
+        distance: Distance from the origin.
+
+    Returns:
+        FloatTensor of shape (4, 4).
+    """
+    azimuth, elevation, distance = float(azimuth), float(elevation), float(distance)
+
+    az_rad = -math.pi * azimuth / 180.0
+    el_rad = -math.pi * elevation / 180.0
+    sa = math.sin(az_rad)
+    ca = math.cos(az_rad)
+    se = math.sin(el_rad)
+    ce = math.cos(el_rad)
+    R_world2obj = torch.tensor(
+        [[ca * ce, sa * ce, -se], [-sa, ca, 0], [ca * se, sa * se, ce]]
+    )
+    R_obj2cam = torch.tensor([[0.0, 1.0, 0.0], [0.0, 0.0, 1.0], [1.0, 0.0, 0.0]])
+    R_world2cam = R_obj2cam.mm(R_world2obj)
+    cam_location = torch.tensor([[distance, 0, 0]]).t()
+    T_world2cam = -(R_obj2cam.mm(cam_location))
+    RT = torch.cat([R_world2cam, T_world2cam], dim=1)
+    RT = torch.cat([RT, torch.tensor([[0.0, 0, 0, 1]])])
+
+    # Georgia: For some reason I cannot fathom, when Blender loads a .obj file it
+    # rotates the model 90 degrees about the x axis. To compensate for this quirk we
+    # roll that rotation into the extrinsic matrix here
+    rot = torch.tensor([[1, 0, 0, 0], [0, 0, -1, 0], [0, 1, 0, 0], [0, 0, 0, 1]])
+    RT = RT.mm(rot.to(RT))
+
+    return RT

tests/test_r2n2.py

@@ -93,10 +93,18 @@ class TestR2N2(TestCaseMixin, unittest.TestCase):
         self.assertEqual(faces.ndim, 2)
         self.assertEqual(faces.shape[-1], 3)
 
+        # Check that the intrinsic matrix and extrinsic matrix have the
+        # correct shapes.
+        self.assertEqual(r2n2_obj["R"].shape[0], 24)
+        self.assertEqual(r2n2_obj["R"].shape[1:], (3, 3))
+        self.assertEqual(r2n2_obj["T"].ndim, 2)
+        self.assertEqual(r2n2_obj["T"].shape[1], 3)
+        self.assertEqual(r2n2_obj["K"].ndim, 3)
+        self.assertEqual(r2n2_obj["K"].shape[1:], (4, 4))
+
         # Check that image batch returned by __getitem__ has the correct shape.
         self.assertEqual(r2n2_obj["images"].shape[0], 24)
-        self.assertEqual(r2n2_obj["images"].shape[1], 137)
-        self.assertEqual(r2n2_obj["images"].shape[2], 137)
+        self.assertEqual(r2n2_obj["images"].shape[1:-1], (137, 137))
         self.assertEqual(r2n2_obj["images"].shape[-1], 3)
         self.assertEqual(r2n2_dataset[39, [21]]["images"].shape[0], 1)
         self.assertEqual(r2n2_dataset[39, torch.tensor([12, 21])]["images"].shape[0], 2)
@@ -113,7 +121,7 @@ class TestR2N2(TestCaseMixin, unittest.TestCase):
         the correct shapes and types are returned.
         """
         # Load dataset in the train split.
-        r2n2_dataset = R2N2("train", SHAPENET_PATH, R2N2_PATH, SPLITS_PATH)
+        r2n2_dataset = R2N2("val", SHAPENET_PATH, R2N2_PATH, SPLITS_PATH)
 
         # Randomly retrieve several objects from the dataset and collate them.
         collated_meshes = collate_batched_meshes(
@@ -147,6 +155,9 @@ class TestR2N2(TestCaseMixin, unittest.TestCase):
         self.assertEqual(object_batch["mesh"].verts_padded().shape[0], batch_size)
         self.assertEqual(object_batch["mesh"].faces_padded().shape[0], batch_size)
         self.assertEqual(object_batch["images"].shape[0], batch_size)
+        self.assertEqual(object_batch["R"].shape[0], batch_size)
+        self.assertEqual(object_batch["T"].shape[0], batch_size)
+        self.assertEqual(object_batch["K"].shape[0], batch_size)
 
     def test_catch_render_arg_errors(self):
         """
@@ -166,6 +177,13 @@ class TestR2N2(TestCaseMixin, unittest.TestCase):
             r2n2_dataset.render(idxs=[1000000])
         self.assertTrue("are out of bounds" in str(err.exception))
 
+        blend_cameras = BlenderCamera(
+            R=torch.rand((3, 3, 3)), T=torch.rand((3, 3)), K=torch.rand((3, 4, 4))
+        )
+        with self.assertRaises(ValueError) as err:
+            r2n2_dataset.render(idxs=[10, 11], cameras=blend_cameras)
+        self.assertTrue("Mismatch between batch dims" in str(err.exception))
+
     def test_render_r2n2(self):
         """
         Test rendering objects from R2N2 selected both by indices and model_ids.
@@ -279,3 +297,44 @@ class TestR2N2(TestCaseMixin, unittest.TestCase):
         C = cam.get_camera_center()
         C_ = -torch.bmm(R, T[:, :, None])[:, :, 0]
         self.assertTrue(torch.allclose(C, C_, atol=1e-05))
+
+    def test_render_by_r2n2_calibration(self):
+        """
+        Test rendering R2N2 models with calibration matrices from R2N2's own Blender
+        in batches.
+        """
+        # Set up device and seed for random selections.
+        device = torch.device("cuda:0")
+        torch.manual_seed(39)
+
+        # Load dataset in the train split.
+        r2n2_dataset = R2N2("train", SHAPENET_PATH, R2N2_PATH, SPLITS_PATH)
+        model_idxs = torch.randint(1000, (2,)).tolist()
+        view_idxs = torch.randint(24, (2,)).tolist()
+        raster_settings = RasterizationSettings(image_size=512)
+        lights = PointLights(
+            location=torch.tensor([0.0, 1.0, -2.0], device=device)[None],
+            # TODO(nikhilar): debug the source of the discrepancy in two images when
+            # rendering on GPU.
+            diffuse_color=((0, 0, 0),),
+            specular_color=((0, 0, 0),),
+            device=device,
+        )
+        r2n2_batch = r2n2_dataset.render(
+            idxs=model_idxs,
+            view_idxs=view_idxs,
+            device=device,
+            raster_settings=raster_settings,
+            lights=lights,
+        )
+        for idx in range(4):
+            r2n2_batch_rgb = r2n2_batch[idx, ..., :3].squeeze().cpu()
+            if DEBUG:
+                Image.fromarray((r2n2_batch_rgb.numpy() * 255).astype(np.uint8)).save(
+                    DATA_DIR
+                    / ("DEBUG_r2n2_render_with_blender_calibrations_%s.png" % idx)
+                )
+            image_ref = load_rgb_image(
+                "test_r2n2_render_with_blender_calibrations_%s.png" % idx, DATA_DIR
+            )
+            self.assertClose(r2n2_batch_rgb, image_ref, atol=0.05)