Add OpenCV camera conversion; fix bug for camera unified PyTorch3D interface.

Summary: This commit adds a new camera conversion function for OpenCV style parameters to Pulsar parameters to the library. Using this function it addresses a bug reported here: https://fb.workplace.com/groups/629644647557365/posts/1079637302558095, by using the PyTorch3D->OpenCV->Pulsar chain instead of the original direct conversion function. Both conversions are well-tested and an additional test for the full chain has been added, resulting in a more reliable solution requiring less code.

Reviewed By: patricklabatut

Differential Revision: D29322106

fbshipit-source-id: 13df13c2e48f628f75d9f44f19ff7f1646fb7ebd

pytorch3d/renderer/points/pulsar/unified.py

@@ -11,7 +11,7 @@ from typing import Any, Dict, Optional, Tuple, Union
 import torch
 import torch.nn as nn
 
-from ....transforms import matrix_to_rotation_6d
+from ....utils import pulsar_from_cameras_projection
 from ...cameras import (
     FoVOrthographicCameras,
     FoVPerspectiveCameras,
@@ -102,7 +102,7 @@ class PulsarPointsRenderer(nn.Module):
             height=height,
             max_num_balls=max_num_spheres,
             orthogonal_projection=orthogonal_projection,
-            right_handed_system=True,
+            right_handed_system=False,
             n_channels=n_channels,
             **kwargs,
         )
@@ -359,24 +359,28 @@ class PulsarPointsRenderer(nn.Module):
     def _extract_extrinsics(
         self, kwargs, cloud_idx
     ) -> Tuple[torch.Tensor, torch.Tensor]:
+        """
+        Extract the extrinsic information from the kwargs for a specific point cloud.
+
+        Instead of implementing a direct translation from the PyTorch3D to the Pulsar
+        camera model, we chain the two conversions of PyTorch3D->OpenCV and
+        OpenCV->Pulsar for better maintainability (PyTorch3D->OpenCV is maintained and
+        tested by the core PyTorch3D team, whereas OpenCV->Pulsar is maintained and
+        tested by the Pulsar team).
+        """
         # Shorthand:
         cameras = self.rasterizer.cameras
         R = kwargs.get("R", cameras.R)[cloud_idx]
         T = kwargs.get("T", cameras.T)[cloud_idx]
-        norm_mat = torch.tensor(
-            [[1.0, 0.0, 0.0], [0.0, 1.0, 0.0], [0.0, 0.0, -1.0]],
-            dtype=torch.float32,
-            device=R.device,
+        tmp_cams = PerspectiveCameras(
+            R=R.unsqueeze(0), T=T.unsqueeze(0), device=R.device
         )
-        cam_rot = torch.matmul(norm_mat, R[:3, :3][None, ...]).permute((0, 2, 1))
-        norm_mat = torch.tensor(
-            [[-1.0, 0.0, 0.0], [0.0, 1.0, 0.0], [0.0, 0.0, 1.0]],
-            dtype=torch.float32,
-            device=R.device,
+        size_tensor = torch.tensor(
+            [[self.renderer._renderer.height, self.renderer._renderer.width]]
         )
-        cam_rot = torch.matmul(norm_mat, cam_rot)
-        cam_pos = torch.flatten(torch.matmul(cam_rot, T[..., None]))
-        cam_rot = torch.flatten(matrix_to_rotation_6d(cam_rot))
+        pulsar_cam = pulsar_from_cameras_projection(tmp_cams, size_tensor)
+        cam_pos = pulsar_cam[0, :3]
+        cam_rot = pulsar_cam[0, 3:9]
         return cam_pos, cam_rot
 
     def _get_vert_rad(
@@ -547,15 +551,17 @@ class PulsarPointsRenderer(nn.Module):
                 otherargs["bg_col"] = bg_col
             # Go!
             images.append(
-                self.renderer(
-                    vert_pos=vert_pos,
-                    vert_col=vert_col,
-                    vert_rad=vert_rad,
-                    cam_params=cam_params,
-                    gamma=gamma,
-                    max_depth=zfar,
-                    min_depth=znear,
-                    **otherargs,
+                torch.flipud(
+                    self.renderer(
+                        vert_pos=vert_pos,
+                        vert_col=vert_col,
+                        vert_rad=vert_rad,
+                        cam_params=cam_params,
+                        gamma=gamma,
+                        max_depth=zfar,
+                        min_depth=znear,
+                        **otherargs,
+                    )
                 )
             )
         return torch.stack(images, dim=0)

pytorch3d/utils/__init__.py

@@ -7,6 +7,8 @@
 from .camera_conversions import (
     cameras_from_opencv_projection,
     opencv_from_cameras_projection,
+    pulsar_from_opencv_projection,
+    pulsar_from_cameras_projection,
 )
 from .ico_sphere import ico_sphere
 from .torus import torus

pytorch3d/utils/camera_conversions.py

@@ -4,12 +4,16 @@
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
+import logging
 from typing import Tuple
 
 import torch
 
 from ..renderer import PerspectiveCameras
-from ..transforms import so3_exp_map, so3_log_map
+from ..transforms import matrix_to_rotation_6d
+
+
+LOGGER = logging.getLogger(__name__)
 
 
 def cameras_from_opencv_projection(
@@ -54,7 +58,6 @@ def cameras_from_opencv_projection(
     Returns:
         cameras_pytorch3d: A batch of `N` cameras in the PyTorch3D convention.
     """
-
     focal_length = torch.stack([camera_matrix[:, 0, 0], camera_matrix[:, 1, 1]], dim=-1)
     principal_point = camera_matrix[:, :2, 2]
 
@@ -68,7 +71,7 @@ def cameras_from_opencv_projection(
     # For R, T we flip x, y axes (opencv screen space has an opposite
     # orientation of screen axes).
     # We also transpose R (opencv multiplies points from the opposite=left side).
-    R_pytorch3d = R.permute(0, 2, 1)
+    R_pytorch3d = R.clone().permute(0, 2, 1)
     T_pytorch3d = tvec.clone()
     R_pytorch3d[:, :, :2] *= -1
     T_pytorch3d[:, :2] *= -1
@@ -103,20 +106,22 @@ def opencv_from_cameras_projection(
         cameras: A batch of `N` cameras in the PyTorch3D convention.
         image_size: A tensor of shape `(N, 2)` containing the sizes of the images
             (height, width) attached to each camera.
+        return_as_rotmat (bool): If set to True, return the full 3x3 rotation
+            matrices. Otherwise, return an axis-angle vector (default).
 
     Returns:
         R: A batch of rotation matrices of shape `(N, 3, 3)`.
         tvec: A batch of translation vectors of shape `(N, 3)`.
         camera_matrix: A batch of camera calibration matrices of shape `(N, 3, 3)`.
     """
-    R_pytorch3d = cameras.R
-    T_pytorch3d = cameras.T
+    R_pytorch3d = cameras.R.clone()  # pyre-ignore
+    T_pytorch3d = cameras.T.clone()  # pyre-ignore
     focal_pytorch3d = cameras.focal_length
     p0_pytorch3d = cameras.principal_point
-    T_pytorch3d[:, :2] *= -1  # pyre-ignore
-    R_pytorch3d[:, :, :2] *= -1  # pyre-ignore
-    tvec = T_pytorch3d.clone()  # pyre-ignore
-    R = R_pytorch3d.permute(0, 2, 1)  # pyre-ignore
+    T_pytorch3d[:, :2] *= -1
+    R_pytorch3d[:, :, :2] *= -1
+    tvec = T_pytorch3d
+    R = R_pytorch3d.permute(0, 2, 1)
 
     # Retype the image_size correctly and flip to width, height.
     image_size_wh = image_size.to(R).flip(dims=(1,))
@@ -130,3 +135,151 @@ def opencv_from_cameras_projection(
     camera_matrix[:, 0, 0] = focal_length[:, 0]
     camera_matrix[:, 1, 1] = focal_length[:, 1]
     return R, tvec, camera_matrix
+
+
+def pulsar_from_opencv_projection(
+    R: torch.Tensor,
+    tvec: torch.Tensor,
+    camera_matrix: torch.Tensor,
+    image_size: torch.Tensor,
+    znear: float = 0.1,
+) -> torch.Tensor:
+    """
+    Convert OpenCV style camera parameters to Pulsar style camera parameters.
+
+    Note:
+        * Pulsar does NOT support different focal lengths for x and y.
+          For conversion, we use the average of fx and fy.
+        * The Pulsar renderer MUST use a left-handed coordinate system for this
+          mapping to work.
+        * The resulting image will be vertically flipped - which has to be
+          addressed AFTER rendering by the user.
+        * The parameters `R, tvec, camera_matrix` correspond to the outputs
+          of `cv2.decomposeProjectionMatrix`.
+
+    Args:
+        R: A batch of rotation matrices of shape `(N, 3, 3)`.
+        tvec: A batch of translation vectors of shape `(N, 3)`.
+        camera_matrix: A batch of camera calibration matrices of shape `(N, 3, 3)`.
+        image_size: A tensor of shape `(N, 2)` containing the sizes of the images
+            (height, width) attached to each camera.
+        znear (float): The near clipping value to use for Pulsar.
+
+    Returns:
+        cameras_pulsar: A batch of `N` Pulsar camera vectors in the Pulsar
+            convention `(N, 13)` (3 translation, 6 rotation, focal_length, sensor_width,
+            c_x, c_y).
+    """
+    assert len(camera_matrix.size()) == 3, "This function requires batched inputs!"
+    assert len(R.size()) == 3, "This function requires batched inputs!"
+    assert len(tvec.size()) in (2, 3), "This function reuqires batched inputs!"
+
+    # Validate parameters.
+    image_size_wh = image_size.to(R).flip(dims=(1,))
+    assert torch.all(
+        image_size_wh > 0
+    ), "height and width must be positive but min is: %s" % (
+        str(image_size_wh.min().item())
+    )
+    assert (
+        camera_matrix.size(1) == 3 and camera_matrix.size(2) == 3
+    ), "Incorrect camera matrix shape: expected 3x3 but got %dx%d" % (
+        camera_matrix.size(1),
+        camera_matrix.size(2),
+    )
+    assert (
+        R.size(1) == 3 and R.size(2) == 3
+    ), "Incorrect R shape: expected 3x3 but got %dx%d" % (
+        R.size(1),
+        R.size(2),
+    )
+    if len(tvec.size()) == 2:
+        tvec = tvec.unsqueeze(2)
+    assert (
+        tvec.size(1) == 3 and tvec.size(2) == 1
+    ), "Incorrect tvec shape: expected 3x1 but got %dx%d" % (
+        tvec.size(1),
+        tvec.size(2),
+    )
+    # Check batch size.
+    batch_size = camera_matrix.size(0)
+    assert R.size(0) == batch_size, "Expected R to have batch size %d. Has size %d." % (
+        batch_size,
+        R.size(0),
+    )
+    assert (
+        tvec.size(0) == batch_size
+    ), "Expected tvec to have batch size %d. Has size %d." % (
+        batch_size,
+        tvec.size(0),
+    )
+    # Check image sizes.
+    image_w = image_size_wh[0, 0]
+    image_h = image_size_wh[0, 1]
+    assert torch.all(
+        image_size_wh[:, 0] == image_w
+    ), "All images in a batch must have the same width!"
+    assert torch.all(
+        image_size_wh[:, 1] == image_h
+    ), "All images in a batch must have the same height!"
+    # Focal length.
+    fx = camera_matrix[:, 0, 0].unsqueeze(1)
+    fy = camera_matrix[:, 1, 1].unsqueeze(1)
+    # Check that we introduce less than 1% error by averaging the focal lengths.
+    fx_y = fx / fy
+    if torch.any(fx_y > 1.01) or torch.any(fx_y < 0.99):
+        LOGGER.warning(
+            "Pulsar only supports a single focal lengths. For converting OpenCV "
+            "focal lengths, we average them for x and y directions. "
+            "The focal lengths for x and y you provided differ by more than 1%, "
+            "which means this could introduce a noticeable error."
+        )
+    f = (fx + fy) / 2
+    # Normalize f into normalized device coordinates.
+    focal_length_px = f / image_w
+    # Transfer into focal_length and sensor_width.
+    focal_length = torch.tensor([znear - 1e-5], dtype=torch.float32, device=R.device)
+    focal_length = focal_length[None, :].repeat(batch_size, 1)
+    sensor_width = focal_length / focal_length_px
+    # Principal point.
+    cx = camera_matrix[:, 0, 2].unsqueeze(1)
+    cy = camera_matrix[:, 1, 2].unsqueeze(1)
+    # Transfer principal point offset into centered offset.
+    cx = -(cx - image_w / 2)
+    cy = cy - image_h / 2
+    # Concatenate to final vector.
+    param = torch.cat([focal_length, sensor_width, cx, cy], dim=1)
+    R_trans = R.permute(0, 2, 1)
+    cam_pos = -torch.bmm(R_trans, tvec).squeeze(2)
+    cam_rot = matrix_to_rotation_6d(R_trans)
+    cam_params = torch.cat([cam_pos, cam_rot, param], dim=1)
+    return cam_params
+
+
+def pulsar_from_cameras_projection(
+    cameras: PerspectiveCameras,
+    image_size: torch.Tensor,
+) -> torch.Tensor:
+    """
+    Convert PyTorch3D `PerspectiveCameras` to Pulsar style camera parameters.
+
+    Note:
+        * Pulsar does NOT support different focal lengths for x and y.
+          For conversion, we use the average of fx and fy.
+        * The Pulsar renderer MUST use a left-handed coordinate system for this
+          mapping to work.
+        * The resulting image will be vertically flipped - which has to be
+          addressed AFTER rendering by the user.
+
+    Args:
+        cameras: A batch of `N` cameras in the PyTorch3D convention.
+        image_size: A tensor of shape `(N, 2)` containing the sizes of the images
+            (height, width) attached to each camera.
+
+    Returns:
+        cameras_pulsar: A batch of `N` Pulsar camera vectors in the Pulsar
+            convention `(N, 13)` (3 translation, 6 rotation, focal_length, sensor_width,
+            c_x, c_y).
+    """
+    opencv_R, opencv_T, opencv_K = opencv_from_cameras_projection(cameras, image_size)
+    return pulsar_from_opencv_projection(opencv_R, opencv_T, opencv_K, image_size)

tests/test_camera_conversions.py

@@ -12,10 +12,12 @@ import numpy as np
 import torch
 from common_testing import TestCaseMixin, get_tests_dir
 from pytorch3d.ops import eyes
+from pytorch3d.renderer.points.pulsar import Renderer as PulsarRenderer
 from pytorch3d.transforms import so3_exp_map, so3_log_map
 from pytorch3d.utils import (
     cameras_from_opencv_projection,
     opencv_from_cameras_projection,
+    pulsar_from_opencv_projection,
 )
 
 
@@ -111,6 +113,9 @@ class TestCameraConversions(TestCaseMixin, unittest.TestCase):
             [105.0, 105.0],
             [120.0, 120.0],
         ]
+        # These values are in y, x format, but they should be in x, y format.
+        # The tests work like this because they only test for consistency,
+        # but this format is misleading.
         principal_point = [
             [240, 320],
             [240.5, 320.3],
@@ -160,3 +165,80 @@ class TestCameraConversions(TestCaseMixin, unittest.TestCase):
         self.assertClose(R, R_i)
         self.assertClose(tvec, tvec_i)
         self.assertClose(camera_matrix, camera_matrix_i)
+
+    def test_pulsar_conversion(self):
+        """
+        Tests that the cameras converted from opencv to pulsar convention
+        return correct projections of random 3D points. The check is done
+        against a set of results precomputed using `cv2.projectPoints` function.
+        """
+        image_size = [[480, 640]]
+        R = [
+            [
+                [1.0, 0.0, 0.0],
+                [0.0, 1.0, 0.0],
+                [0.0, 0.0, 1.0],
+            ],
+            [
+                [0.1968, -0.6663, -0.7192],
+                [0.7138, -0.4055, 0.5710],
+                [-0.6721, -0.6258, 0.3959],
+            ],
+        ]
+        tvec = [
+            [10.0, 10.0, 3.0],
+            [-0.0, -0.0, 20.0],
+        ]
+        focal_length = [
+            [100.0, 100.0],
+            [10.0, 10.0],
+        ]
+        principal_point = [
+            [320, 240],
+            [320, 240],
+        ]
+
+        principal_point, focal_length, R, tvec, image_size = [
+            torch.FloatTensor(x)
+            for x in (principal_point, focal_length, R, tvec, image_size)
+        ]
+        camera_matrix = eyes(dim=3, N=2)
+        camera_matrix[:, 0, 0] = focal_length[:, 0]
+        camera_matrix[:, 1, 1] = focal_length[:, 1]
+        camera_matrix[:, :2, 2] = principal_point
+        rvec = so3_log_map(R)
+        pts = torch.tensor(
+            [[[0.0, 0.0, 120.0]], [[0.0, 0.0, 120.0]]], dtype=torch.float32
+        )
+        radii = torch.tensor([[1e-5], [1e-5]], dtype=torch.float32)
+        col = torch.zeros((2, 1, 1), dtype=torch.float32)
+
+        # project the 3D points with the opencv projection function
+        pts_proj_opencv = cv2_project_points(pts, rvec, tvec, camera_matrix)
+        pulsar_cam = pulsar_from_opencv_projection(
+            R, tvec, camera_matrix, image_size, znear=100.0
+        )
+        pulsar_rend = PulsarRenderer(
+            640, 480, 1, right_handed_system=False, n_channels=1
+        )
+        rendered = torch.flip(
+            pulsar_rend(
+                pts,
+                col,
+                radii,
+                pulsar_cam,
+                1e-5,
+                max_depth=150.0,
+                min_depth=100.0,
+            ),
+            dims=(1,),
+        )
+        for batch_id in range(2):
+            point_pos = torch.where(rendered[batch_id] == rendered[batch_id].min())
+            point_pos = point_pos[1][0], point_pos[0][0]
+            self.assertLess(
+                torch.abs(point_pos[0] - pts_proj_opencv[batch_id, 0, 0]), 2
+            )
+            self.assertLess(
+                torch.abs(point_pos[1] - pts_proj_opencv[batch_id, 0, 1]), 2
+            )