-
Notifications
You must be signed in to change notification settings - Fork 2
/
AgentPrediction_config_Baseline.yaml
133 lines (104 loc) · 3.56 KB
/
AgentPrediction_config_Baseline.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
120
121
122
123
124
125
126
127
128
129
130
131
132
# Config format schema number
format_version: 4
###################
## Model options
model_params:
version: "Base"
# Estimate yaw in addition to X and Y
use_angle: False
# If true all history image channels are computed as a fadded image.
# The most recent frames have higher intensity an it decays into the past
use_fading: False
# Base model for image feature extraction
# Note: Each model has its own preprocessing function, please select the right one...
# base_image_model: "ResNet50"
# base_image_preprocess: "resnet.preprocess_input"
base_image_model: "MobileNetV2"
base_image_preprocess: "mobilenet_v2.preprocess_input"
# Number of fully connected layers used to generate the output coordinates (not counting last)
num_path_decode_fc_layers: 1
# Number of units in the FC layers used to generate the output coordinates
path_decode_fc_units: [4096]
# Activation of the FC layers used to generate the output coordinates
path_decode_fc_activation: 'relu'
# Number of history frames
history_num_frames: 10
# Steps between history frames
history_step_size: 1
# History sampling frequency
history_delta_time: 0.1
# Number of future frames
future_num_frames: 50
# Step size of future frames
future_step_size: 1
# Future steps sampling
future_delta_time: 0.1
###################
## Training options
training_params:
# Learning rate list and epochs conforming a learning rate schedule
gen_lr_list: [0.001, 0.0001, 0.00001, 0.00001]
gen_lr_lims: [ 2, 10, 20, 1000]
# Number of scenes to be used in training
number_of_scenes: 16000
# Whether or not pick the scenes at random
randomize_scenes: True
# Frames to be read from a given scene
frames_per_scene: 1
# Whether or not pick the frames inside a given scene at random
randomize_frames: True
# Epochs to train, an epoch is a complete iteration over the "number_of_scenes",
# computing "frames_per_scene"x"number_of_scenes" samples.
epochs_train: 1000
# Whether or not retrain input layer of image processing model
retrain_inputs_image_model: True
# Whether or not retrain the whole image processing model
retrain_all_image_model: True
###################
## Input raster parameters
raster_params:
# raster image size [pixels]
raster_size:
- 224
- 224
# raster's spatial resolution [meters per pixel]: the size in the real world one pixel corresponds to.
pixel_size:
- 0.5
- 0.5
# From 0 to 1 per axis, [0.5,0.5] would show the ego centered in the image.
ego_center:
- 0.25
- 0.5
map_type: "py_semantic"
# map_type: "py_satellite"
# the keys are relative to the dataset environment variable
satellite_map_key: "aerial_map/aerial_map.png"
semantic_map_key: "semantic_map/semantic_map.pb"
dataset_meta_key: "meta.json"
# e.g. 0.0 include every obstacle, 0.5 show those obstacles with >0.5 probability of being
# one of the classes we care about (cars, bikes, peds, etc.), >=1.0 filter all other agents.
filter_agents_threshold: 0.5
# whether to completely disable traffic light faces in the semantic rasterizer
disable_traffic_light_faces: False
###################
## Data loader options
sample_data_loader:
key: "scenes/sample.zarr"
batch_size: 64
shuffle: False
num_workers: 4
train_data_loader:
key: "scenes/train.zarr"
batch_size: 32
shuffle: True
num_workers: 4
val_data_loader:
key: "scenes/validate.zarr"
batch_size: 32
shuffle: False
num_workers: 4
test_data_loader:
key: "scenes/test.zarr"
batch_size: 32
shuffle: False
num_workers: 4