Spaces:
Runtime error
Runtime error
File size: 4,870 Bytes
2df809d |
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 |
import os.path as osp
import cv2
import numpy as np
import itertools
import os
import sys
sys.path.append(osp.join(osp.dirname(__file__), "..", ".."))
from tqdm import tqdm
from dust3r.datasets.base.base_multiview_dataset import BaseMultiViewDataset
from dust3r.utils.image import imread_cv2
class MVS_Synth_Multi(BaseMultiViewDataset):
def __init__(self, *args, ROOT, **kwargs):
self.ROOT = ROOT
self.video = True
self.is_metric = False
self.max_interval = 4
super().__init__(*args, **kwargs)
self.loaded_data = self._load_data()
def _load_data(self):
self.scenes = os.listdir(self.ROOT)
offset = 0
scenes = []
sceneids = []
scene_img_list = []
images = []
start_img_ids = []
j = 0
for scene in tqdm(self.scenes):
scene_dir = osp.join(self.ROOT, scene)
rgb_dir = osp.join(scene_dir, "rgb")
basenames = sorted(
[f[:-4] for f in os.listdir(rgb_dir) if f.endswith(".jpg")]
)
num_imgs = len(basenames)
cut_off = (
self.num_views if not self.allow_repeat else max(self.num_views // 3, 3)
)
if num_imgs < cut_off:
print(f"Skipping {scene}")
continue
img_ids = list(np.arange(num_imgs) + offset)
start_img_ids_ = img_ids[: num_imgs - cut_off + 1]
start_img_ids.extend(start_img_ids_)
sceneids.extend([j] * num_imgs)
images.extend(basenames)
scenes.append(scene)
scene_img_list.append(img_ids)
# offset groups
offset += num_imgs
j += 1
self.scenes = scenes
self.sceneids = sceneids
self.images = images
self.start_img_ids = start_img_ids
self.scene_img_list = scene_img_list
def __len__(self):
return len(self.start_img_ids)
def get_image_num(self):
return len(self.images)
def _get_views(self, idx, resolution, rng, num_views):
start_id = self.start_img_ids[idx]
all_image_ids = self.scene_img_list[self.sceneids[start_id]]
pos, ordered_video = self.get_seq_from_start_id(
num_views,
start_id,
all_image_ids,
rng,
max_interval=self.max_interval,
video_prob=1.0,
fix_interval_prob=1.0,
)
image_idxs = np.array(all_image_ids)[pos]
views = []
for v, view_idx in enumerate(image_idxs):
scene_id = self.sceneids[view_idx]
scene_dir = osp.join(self.ROOT, self.scenes[scene_id])
rgb_dir = osp.join(scene_dir, "rgb")
depth_dir = osp.join(scene_dir, "depth")
cam_dir = osp.join(scene_dir, "cam")
basename = self.images[view_idx]
# Load RGB image
rgb_image = imread_cv2(osp.join(rgb_dir, basename + ".jpg"))
# Load depthmap
depthmap = np.load(osp.join(depth_dir, basename + ".npy"))
depthmap[~np.isfinite(depthmap)] = 0 # invalid
threshold = (
np.percentile(depthmap[depthmap > 0], 98)
if depthmap[depthmap > 0].size > 0
else 0
)
depthmap[depthmap > threshold] = 0.0
depthmap[depthmap > 1000] = 0.0
cam = np.load(osp.join(cam_dir, basename + ".npz"))
camera_pose = cam["pose"]
intrinsics = cam["intrinsics"]
rgb_image, depthmap, intrinsics = self._crop_resize_if_necessary(
rgb_image, depthmap, intrinsics, resolution, rng=rng, info=view_idx
)
# generate img mask and raymap mask
img_mask, ray_mask = self.get_img_and_ray_masks(
self.is_metric, v, rng, p=[0.8, 0.15, 0.05]
)
views.append(
dict(
img=rgb_image,
depthmap=depthmap.astype(np.float32),
camera_pose=camera_pose.astype(np.float32),
camera_intrinsics=intrinsics.astype(np.float32),
dataset="MVS_Synth",
label=self.scenes[scene_id] + "_" + basename,
instance=osp.join(rgb_dir, basename + ".jpg"),
is_metric=self.is_metric,
is_video=ordered_video,
quantile=np.array(1.0, dtype=np.float32),
img_mask=img_mask,
ray_mask=ray_mask,
camera_only=False,
depth_only=False,
single_view=False,
reset=False,
)
)
assert len(views) == num_views
return views
|