-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathparams.yaml
119 lines (99 loc) · 3.4 KB
/
params.yaml
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
###############################
#######General#################
###############################
#the amount of images supposed to be processed
n_images: 800
#starting number (usually 0)
n_start: 200
#only process all n-th pictures
skip: 15
#determines how big the voxels are supposed to be -> if you're running out of memory just increase the size
voxel_size: 0.02
filter_interval: -1 # do not apply median filter
#size of the processed images
img_size: !!python/tuple [640, 480]
#0: only using dynamic mask generator
#1: only using file mask generator (only works if all masks can be found at mask_path)
#2: using both (uses more gpu power than 1)
#3: using ScanNet ground truth, requires gt_instance_path & gt_label_path set correctly
segment_mode: 2
#############################################
#########Preprocessor Parameters#############
#############################################
semantic_masks:
#masks with a score beneath this threshold get skipped
score_threshold: 0.2
# all values below that in a mask get set to zero
mask_threshold: 0.3
#threshold for evaluating the median and superpixel stuff of the mask
rcnn_threshold: 150
#threshold for the score in the class image
rcnn_threshold_2: 40
#determines if we want to use depth images to improve mask scores (decreases performance by a factor ~x1.5)
use_depth: False
#falloff for depth score
depth_f: 5
#determines if we want to use superpixel or not. (decreases performance by a factor ~x5)
#both of the features above together decrease the performance of my laptop from 1.5fps to 0.2fps
use_sp: False
fast_slic_comp: 500 # number of components for fast_slic
prefetch_size: 1
save_masks: False
instance_generation:
#if True will merge all classes at the end
combine_instances: True
#amount of voxels which have to touch for 2 instances to be merged
hit_count: 500
#threshold for merging 2 instances
instance_threshold: 0.8
#if the amount of available instances runs low we use a lower hit_count to free up some instance space
hit_count_low: 200
#########
##Paths##
###########################################################################
#Model
model_path: "src/MaskRCNNImpl/Models/Resnet_40_trained.pth"
labels_path: "Dataset/labels_40.csv"
camera_intrinsics_path : "Dataset/camera-intrinsics.txt"
mask_path: "Dataset/masks"
rgb_path: "Dataset/color"
d_path: "Dataset/depth"
pose_path: "Dataset/pose"
output_path: "output"
#
### Only needed when segment_mode is set to 3
gt_instance_path: "Dataset/groundtruth/instance-filt"
gt_label_path: "Dataset/groundtruth/label-filt"
###################################
##########Debug####################
###################################
debug:
extract_meshes_for_instance:
enabled: False
class: 40
monitor_progress:
enabled: False
volume_images: False
path: "Dataset/progress"
monitor_instances:
enabled: False
classes: 3 #-1 for all otherwise specific class number
combine: False # combine instances after every frame
export_class_image:
enabled: False
path: "Dataset/progress"
ignore_score: True
export_masks:
enabled: False
path: "Dataset/progress"
fusion:
enabled: False
classes: True
instances: True
duplicate_check: True
instance_generator:
enabled: False
class_number: 3 # class number to debug or -1 if all classes
groundtruth_mask:
save: False
path: "E:/new_scene/scene0003_00"