File size: 4,792 Bytes
864ec44 |
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 |
# Last modified: 2024-02-08
#
# Copyright 2023 Bingxin Ke, ETH Zurich. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# --------------------------------------------------------------------------
# If you find this code useful, we kindly ask you to cite our paper in your work.
# Please find bibtex at: https://github.com/prs-eth/Marigold#-citation
# If you use or adapt this code, please attribute to https://github.com/prs-eth/marigold.
# More information about the method can be found at https://marigoldmonodepth.github.io
# --------------------------------------------------------------------------
import torch
from .eval_base_dataset import DepthFileNameMode, EvaluateBaseDataset
class KITTIDataset(EvaluateBaseDataset):
def __init__(
self,
kitti_bm_crop, # Crop to KITTI benchmark size
valid_mask_crop, # Evaluation mask. [None, garg or eigen]
**kwargs,
) -> None:
super().__init__(
# KITTI data parameter
min_depth=1e-5,
max_depth=80,
has_filled_depth=False,
name_mode=DepthFileNameMode.id,
**kwargs,
)
self.kitti_bm_crop = kitti_bm_crop
self.valid_mask_crop = valid_mask_crop
assert self.valid_mask_crop in [
None,
"garg", # set evaluation mask according to Garg ECCV16
"eigen", # set evaluation mask according to Eigen NIPS14
], f"Unknown crop type: {self.valid_mask_crop}"
# Filter out empty depth
self.filenames = [f for f in self.filenames if "None" != f[1]]
def _read_depth_file(self, rel_path):
depth_in = self._read_image(rel_path)
# Decode KITTI depth
depth_decoded = depth_in / 256.0
return depth_decoded
def _load_rgb_data(self, rgb_rel_path):
rgb_data = super()._load_rgb_data(rgb_rel_path)
if self.kitti_bm_crop:
rgb_data = {k: self.kitti_benchmark_crop(v) for k, v in rgb_data.items()}
return rgb_data
def _load_depth_data(self, depth_rel_path, filled_rel_path):
depth_data = super()._load_depth_data(depth_rel_path, filled_rel_path)
if self.kitti_bm_crop:
depth_data = {
k: self.kitti_benchmark_crop(v) for k, v in depth_data.items()
}
return depth_data
@staticmethod
def kitti_benchmark_crop(input_img):
"""
Crop images to KITTI benchmark size
Args:
`input_img` (torch.Tensor): Input image to be cropped.
Returns:
torch.Tensor:Cropped image.
"""
KB_CROP_HEIGHT = 352
KB_CROP_WIDTH = 1216
height, width = input_img.shape[-2:]
top_margin = int(height - KB_CROP_HEIGHT)
left_margin = int((width - KB_CROP_WIDTH) / 2)
if 2 == len(input_img.shape):
out = input_img[
top_margin : top_margin + KB_CROP_HEIGHT,
left_margin : left_margin + KB_CROP_WIDTH,
]
elif 3 == len(input_img.shape):
out = input_img[
:,
top_margin : top_margin + KB_CROP_HEIGHT,
left_margin : left_margin + KB_CROP_WIDTH,
]
return out
def _get_valid_mask(self, depth: torch.Tensor):
# reference: https://github.com/cleinc/bts/blob/master/pytorch/bts_eval.py
valid_mask = super()._get_valid_mask(depth) # [1, H, W]
if self.valid_mask_crop is not None:
eval_mask = torch.zeros_like(valid_mask.squeeze()).bool()
gt_height, gt_width = eval_mask.shape
if "garg" == self.valid_mask_crop:
eval_mask[
int(0.40810811 * gt_height) : int(0.99189189 * gt_height),
int(0.03594771 * gt_width) : int(0.96405229 * gt_width),
] = 1
elif "eigen" == self.valid_mask_crop:
eval_mask[
int(0.3324324 * gt_height) : int(0.91351351 * gt_height),
int(0.0359477 * gt_width) : int(0.96405229 * gt_width),
] = 1
eval_mask.reshape(valid_mask.shape)
valid_mask = torch.logical_and(valid_mask, eval_mask)
return valid_mask
|