Benchmark Cameras

Summary: Address comments to add benchmarkings for cameras and the new fisheye cameras. The dependency functions in test_cameras have been updated in Diff 1. The following two snapshots show benchmarking results.

Reviewed By: kjchalup

Differential Revision: D38991914

fbshipit-source-id: 51fe9bb7237543e4ee112c9f5068a4cf12a9d482
This commit is contained in:
Jiali Duan 2022-08-28 11:43:46 -07:00 committed by Facebook GitHub Bot
parent 2283c292a9
commit 03562d87f5
2 changed files with 105 additions and 6 deletions

View File

@ -0,0 +1,59 @@
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the BSD-style license found in the
# LICENSE file in the root directory of this source tree.
import itertools
from fvcore.common.benchmark import benchmark
from tests.test_cameras import TestCamerasCommon
def _setUp():
case_grid = {
"cam_type": [
"OpenGLOrthographicCameras",
"OpenGLPerspectiveCameras",
"SfMOrthographicCameras",
"SfMPerspectiveCameras",
"FoVOrthographicCameras",
"FoVPerspectiveCameras",
"OrthographicCameras",
"PerspectiveCameras",
"FishEyeCameras",
],
"batch_size": [1, 10],
"num_points": [10, 100],
"device": ["cpu", "cuda:0"],
}
test_cases = itertools.product(*case_grid.values())
kwargs_list = [dict(zip(case_grid.keys(), case)) for case in test_cases]
return kwargs_list
def _bm_cameras_project() -> None:
kwargs_list = _setUp()
benchmark(
TestCamerasCommon.transform_points,
"TEST_TRANSFORM_POINTS",
kwargs_list,
)
def _bm_cameras_unproject() -> None:
kwargs_list = _setUp()
benchmark(
TestCamerasCommon.unproject_points,
"TEST_UNPROJECT_POINTS",
kwargs_list,
)
def bm_cameras() -> None:
_bm_cameras_project()
_bm_cameras_unproject()
if __name__ == "__main__":
bm_cameras()

View File

@ -36,6 +36,7 @@ import unittest
import numpy as np
import torch
from pytorch3d.common.datatypes import Device
from pytorch3d.renderer.camera_utils import join_cameras_as_batch
from pytorch3d.renderer.cameras import (
camera_position_from_spherical_angles,
@ -149,14 +150,17 @@ def ndc_to_screen_points_naive(points, imsize):
def init_random_cameras(
cam_type: typing.Type[CamerasBase], batch_size: int, random_z: bool = False
cam_type: typing.Type[CamerasBase],
batch_size: int,
random_z: bool = False,
device: Device = "cpu",
):
cam_params = {}
T = torch.randn(batch_size, 3) * 0.03
if not random_z:
T[:, 2] = 4
R = so3_exp_map(torch.randn(batch_size, 3) * 3.0)
cam_params = {"R": R, "T": T}
cam_params = {"R": R, "T": T, "device": device}
if cam_type in (OpenGLPerspectiveCameras, OpenGLOrthographicCameras):
cam_params["znear"] = torch.rand(batch_size) * 10 + 0.1
cam_params["zfar"] = torch.rand(batch_size) * 4 + 1 + cam_params["znear"]
@ -613,15 +617,33 @@ class TestCamerasCommon(TestCaseMixin, unittest.TestCase):
self.assertTrue(torch.allclose(xyz_unproj, matching_xyz, atol=1e-4))
@staticmethod
def unproject_points(cam_type, batch_size=50, num_points=100):
def unproject_points(
cam_type, batch_size=50, num_points=100, device: Device = "cpu"
):
"""
Checks that an unprojection of a randomly projected point cloud
stays the same.
"""
if device == "cuda":
device = torch.device("cuda:0")
else:
device = torch.device("cpu")
str2cls = { # noqa
"OpenGLOrthographicCameras": OpenGLOrthographicCameras,
"OpenGLPerspectiveCameras": OpenGLPerspectiveCameras,
"SfMOrthographicCameras": SfMOrthographicCameras,
"SfMPerspectiveCameras": SfMPerspectiveCameras,
"FoVOrthographicCameras": FoVOrthographicCameras,
"FoVPerspectiveCameras": FoVPerspectiveCameras,
"OrthographicCameras": OrthographicCameras,
"PerspectiveCameras": PerspectiveCameras,
"FishEyeCameras": FishEyeCameras,
}
def run_cameras():
# init the cameras
cameras = init_random_cameras(cam_type, batch_size)
cameras = init_random_cameras(str2cls[cam_type], batch_size, device=device)
# xyz - the ground truth point cloud
xyz = torch.randn(num_points, 3) * 0.3
xyz = cameras.unproject_points(xyz, scaled_depth_input=True)
@ -666,15 +688,33 @@ class TestCamerasCommon(TestCaseMixin, unittest.TestCase):
self.assertClose(xyz_project_screen, xyz_project_screen_naive, atol=1e-4)
@staticmethod
def transform_points(cam_type, batch_size=50, num_points=100):
def transform_points(
cam_type, batch_size=50, num_points=100, device: Device = "cpu"
):
"""
Checks that an unprojection of a randomly projected point cloud
stays the same.
"""
if device == "cuda":
device = torch.device("cuda:0")
else:
device = torch.device("cpu")
str2cls = { # noqa
"OpenGLOrthographicCameras": OpenGLOrthographicCameras,
"OpenGLPerspectiveCameras": OpenGLPerspectiveCameras,
"SfMOrthographicCameras": SfMOrthographicCameras,
"SfMPerspectiveCameras": SfMPerspectiveCameras,
"FoVOrthographicCameras": FoVOrthographicCameras,
"FoVPerspectiveCameras": FoVPerspectiveCameras,
"OrthographicCameras": OrthographicCameras,
"PerspectiveCameras": PerspectiveCameras,
"FishEyeCameras": FishEyeCameras,
}
def run_cameras():
# init the cameras
cameras = init_random_cameras(cam_type, batch_size)
cameras = init_random_cameras(str2cls[cam_type], batch_size, device=device)
# xyz - the ground truth point cloud
xy = torch.randn(num_points, 2) * 2.0 - 1.0
z = torch.randn(num_points, 1) * 3.0 + 1.0