-
Notifications
You must be signed in to change notification settings - Fork 0
/
iqf-usecase.py
executable file
·67 lines (61 loc) · 2.09 KB
/
iqf-usecase.py
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
#! /opt/conda/envs/iqf/bin/python
import sys
import os
#sys.path.append('OBBDetection')
os.chdir('OBBDetection')
from iquaflow.datasets import DSWrapper
from iquaflow.experiments import ExperimentSetup
from iquaflow.experiments.task_execution import PythonScriptTaskExecution
from custom_iqf import (
DSModifierResize,
DSModifier_jpg,
DSModifier_quant
)
experiment = ExperimentSetup(
experiment_name = "iq-dota-obb-use-case",
task_instance = PythonScriptTaskExecution(
model_script_path = '../custom_train.py' ,
tmp_dir='../remove'
),
ref_dsw_train = DSWrapper(
data_path=f'data/split_ss_dota1_0/train',
data_input=f'data/split_ss_dota1_0/train/images',
mask_annotations_dir=os.path.join('data/split_ss_dota1_0/train','annfiles')
),
ref_dsw_val = DSWrapper(
data_path=f'data/split_ss_dota1_0/val',
data_input=f'data/split_ss_dota1_0/val/images',
mask_annotations_dir=os.path.join('data/split_ss_dota1_0/val','annfiles')
),
ref_dsw_test = DSWrapper(
data_path=f'data/split_ss_dota1_0/test',
data_input=f'data/split_ss_dota1_0/test/images',
mask_annotations_dir=os.path.join('data/split_ss_dota1_0/test','annfiles')
),
ds_modifiers_list = [
# DSModifier_jpg(params={"quality": quality}) for quality in range(90,101,2)
] + [
DSModifier_jpg(params={"quality": quality}) for quality in range(10,100,10)
] + [
DSModifier_quant(params={"bits": bits}) for bits in range(1,9)
] + [
DSModifierResize(params={"scaleperc": perc}) for perc in range(10,110,10)
],
repetitions = 1,
mlflow_monitoring = True,
cloud_options = {
'tracking_uri':'file:///work/mlruns'
},
extra_train_params = {
'cu':['0,1,2,3,4,5'],
'ds': ['dota10'],
'model':[
'fcos',
'rcnn',
# 'roitrans'
],
'seed': [98],
}
)
#Execute the experiment
experiment.execute()