File size: 8,291 Bytes
6d69218 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 |
import json
from pathlib import Path
import cv2
import numpy as np
import py360convert
import torch
from scipy.spatial.transform import Rotation as R
from tqdm import tqdm
def compute_focal_length(image_size, fov_deg):
return (image_size / 2) / np.tan(np.deg2rad(fov_deg) / 2)
class OmniVideoProcessor:
default_params = {
"fx": 320.0,
"fy": 320.0,
"cx": 320.0,
"cy": 320.0,
"height": 640,
"width": 640,
"fov_h": 90,
"fov_v": 90,
"frame_interval": 24,
"num_steps_yaw": 4,
"pitches_deg": [-35.0, 35.0],
"views": {
"pitch_35_yaw_0": (35, 0),
"pitch_35_yaw_90": (35, 60),
"pitch_35_yaw_-90": (35, -90),
"pitch_35_yaw_180": (35, 180),
"pitch_-35_yaw_0": (-35, 0),
"pitch_-35_yaw_90": (-35, 60),
"pitch_-35_yaw_-90": (-35, -90),
"pitch_-35_yaw_180": (-35, 180),
},
}
def __init__(self, params={}):
self.params = params if params else self.default_params.copy()
self.ref_sensor = list(self.params["views"].keys())[0]
def set_params(self, params):
self.params = params
def process_video(self, video_or_path, output_dir):
output_dir = Path(output_dir)
output_dir.mkdir(parents=True, exist_ok=True)
if isinstance(video_or_path, str):
video_file = Path(video_or_path)
video = cv2.VideoCapture(str(video_file))
if not video.isOpened():
raise IOError(f"Cannot open video file: {video_file}")
pano_images = self._extract_frames(video, output_dir)
video.release()
elif isinstance(video_or_path, torch.Tensor) or isinstance(video_or_path, np.ndarray):
pano_images = self._extract_frames_torch(video_or_path)
else:
raise ValueError("video_or_path must be a string or Path object")
pinhole_images_data = self._generate_pinhole_images(pano_images, output_dir)
return pano_images, pinhole_images_data
def _extract_frames(self, video, output_dir):
frame_count = int(video.get(cv2.CAP_PROP_FRAME_COUNT))
pano_images = []
for frame_idx in tqdm(range(frame_count), desc="Extracting Frames"):
ret, frame = video.read()
if not ret:
break
if frame_idx % self.params["frame_interval"] == 0:
pano_images.append({"image": frame, "idx": frame_idx})
return pano_images
def _extract_frames_torch(self, video_tensor):
if not isinstance(video_tensor, torch.Tensor):
raise ValueError("video_tensor must be a torch.Tensor")
pano_images = []
num_frames = video_tensor.shape[0]
for frame_idx in tqdm(range(num_frames), desc="Extracting Frames"):
if frame_idx % self.params["frame_interval"] == 0:
# Assuming video_tensor is normalized [0, 1], rgb mode
img = video_tensor[frame_idx].numpy() * 255.0
img = img.astype(np.uint8) # Convert to uint8
pano_images.append({"image": img, "idx": frame_idx})
return pano_images
def _generate_pinhole_images(self, pano_images, output_dir):
output_pinhole_dir = output_dir / "pinhole_images" / "images"
output_pinhole_dir.mkdir(parents=True, exist_ok=True)
pinhole_data = []
camera_params_list = []
camera_rig_params = {}
pinhole_views = []
for pano_info in tqdm(pano_images, desc="Generating Pinhole Views"):
pano_idx, pano_image = pano_info["idx"], pano_info["image"]
for view_name, (pitch, yaw) in self.params["views"].items():
pinhole_image = self._convert_to_pinhole(pano_image, pitch, yaw)
save_dir = output_pinhole_dir / view_name
save_dir.mkdir(parents=True, exist_ok=True)
save_path = save_dir / f"{pano_idx:06d}.jpg"
cv2.imwrite(str(save_path), pinhole_image)
h, w = pinhole_image.shape[:2]
pinhole_views.append(
{
"image": pinhole_image,
"pano_index": pano_idx,
"view_name": view_name,
"pitch": pitch,
"yaw": yaw,
"width": w,
"height": h,
"save_path": str(save_path),
}
)
pinhole_data.append((pano_idx, view_name, pinhole_image, str(save_path)))
is_ref = view_name == self.ref_sensor
cam_params = self._create_camera_params(
save_path, pano_idx, view_name, pitch, yaw, is_ref
)
camera_params_list.append(cam_params)
if view_name not in camera_rig_params:
camera_rig_params[view_name] = {
"image_prefix": view_name,
"yaw": yaw,
"pitch": pitch,
"ref_sensor": is_ref,
}
self._save_camera_params(
camera_params_list,
output_dir / "pinhole_images" / "camera_params.json",
)
self._save_colmap_camera_rig(
camera_rig_params, output_dir / "pinhole_images" / "rig_config.json"
)
return pinhole_views
def _convert_to_pinhole(self, pano_image, pitch, yaw):
return py360convert.e2p(
e_img=pano_image,
fov_deg=(self.params["fov_h"], self.params["fov_v"]),
u_deg=yaw,
v_deg=pitch,
out_hw=(self.params["height"], self.params["width"]),
in_rot_deg=0,
mode="bilinear",
)
def _create_camera_params(
self, save_path: Path, pano_idx, view_name, pitch, yaw, ref_sensor=None
):
fx = compute_focal_length(self.params["width"], self.params["fov_h"])
fy = compute_focal_length(self.params["height"], self.params["fov_v"])
return {
"image_name": save_path.name,
"image_prefix": view_name,
"fx": fx,
"fy": fy,
"cx": self.params["width"] / 2,
"cy": self.params["height"] / 2,
"height": self.params["height"],
"width": self.params["width"],
"fov_h": self.params["fov_h"],
"fov_v": self.params["fov_v"],
"yaw": yaw,
"pitch": pitch,
"pano_index": pano_idx,
"ref_sensor": ref_sensor,
}
def _save_camera_params(self, params, output_file):
with open(output_file, "w") as f:
json.dump(params, f, indent=4)
def _save_colmap_camera_rig(self, camera_rig_params, output_file):
if not self.params["views"]:
return
ref_view_name = list(self.params["views"].keys())[0]
ref_pitch, ref_yaw = self.params["views"][ref_view_name]
# COLMAP: X right, Y down, Z forward. Euler: yaw, pitch, roll
R_ref_world = R.from_euler("yx", [ref_yaw, ref_pitch], degrees=True)
rig_cameras = []
for image_prefix, params in camera_rig_params.items():
R_view_world = R.from_euler("yx", [params["yaw"], params["pitch"]], degrees=True)
R_view_ref = R_view_world.inv() * R_ref_world # Cam from Rig
# Scipy quat (x,y,z,w) -> COLMAP quat (w,x,y,z)
qvec_scipy = R_view_ref.as_quat()
qvec_colmap = [
qvec_scipy[3],
qvec_scipy[0],
qvec_scipy[1],
qvec_scipy[2],
]
cam_entry = {"image_prefix": image_prefix}
if params.get("ref_sensor"):
cam_entry["ref_sensor"] = True
else:
cam_entry["cam_from_rig_rotation"] = qvec_colmap
cam_entry["cam_from_rig_translation"] = [0.0, 0.0, 0.0]
rig_cameras.append(cam_entry)
colmap_rig_config = [{"cameras": rig_cameras}]
with open(output_file, "w") as f:
json.dump(colmap_rig_config, f, indent=4)
|