forked from DeepLabCut/DeepLabCut
-
Notifications
You must be signed in to change notification settings - Fork 0
/
testscript_mobilenets.py
142 lines (111 loc) · 6.55 KB
/
testscript_mobilenets.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
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
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Tue Oct 2 13:56:11 2018
@author: alex
DEVELOPERS:
This script tests various functionalities (creating project ,training, evaluating, outlierextraction, retraining...) in an automatic way.
For that purpose, it trains ResNet and MobileNet briefly on a "fake" dataset.
It should take about 4:15 minutes to run this in a CPU. (incl. downloading the ResNet + MobileNet weights)
It produces nothing of interest scientifically.
"""
task='TEST-multi' # Enter the name of your experiment Task
scorer='Alex' # Enter the name of the experimenter/labeler
import os, subprocess, deeplabcut
from pathlib import Path
import pandas as pd
import numpy as np
def Cuttrainingschedule(path_config_file,shuffle,trainingsetindex=0,initweights='imagenet',lastvalue=10):
cfg=deeplabcut.auxiliaryfunctions.read_config(path_config_file)
posefile=os.path.join(cfg['project_path'],'dlc-models/iteration-'+str(cfg['iteration'])+'/'+ cfg['Task'] + cfg['date'] + '-trainset' + str(int(cfg['TrainingFraction'][trainingsetindex] * 100)) + 'shuffle' + str(shuffle),'train/pose_cfg.yaml')
DLC_config=deeplabcut.auxiliaryfunctions.read_plainconfig(posefile)
DLC_config['save_iters']=lastvalue
DLC_config['display_iters']=1
DLC_config['multi_step']=[[0.001,lastvalue]]
DLC_config['intermediate_supervision']=False #True
if initweights=='previteration':
DLC_config['init_weights']=os.path.join(cfg['project_path'],'dlc-models/iteration-'+str(cfg['iteration']-1)+'/'+ cfg['Task'] + cfg['date'] + '-trainset' + str(int(cfg['TrainingFraction'][trainingsetindex] * 100)) + 'shuffle' + str(shuffle),'train/snapshot-'+str(lastvalue))
print("CHANGING training parameters to end quickly!")
deeplabcut.auxiliaryfunctions.write_plainconfig(posefile,DLC_config)
return
print("Imported DLC!")
basepath=os.path.dirname(os.path.abspath('testscript.py'))
videoname='reachingvideo1'
video=[os.path.join(basepath,'Reaching-Mackenzie-2018-08-30','videos',videoname+'.avi')]
#to test destination folder:
dfolder=basepath
#dfolder=None
augmenter_type='tensorpack' #imgaug'
print("CREATING PROJECT")
path_config_file=deeplabcut.create_new_project(task,scorer,video,copy_videos=True)
cfg=deeplabcut.auxiliaryfunctions.read_config(path_config_file)
cfg['numframes2pick']=5
cfg['pcutoff']=0.01
cfg['TrainingFraction']=[.8]
cfg['skeleton']=[['bodypart1','bodypart2'],['bodypart1','bodypart3']]
deeplabcut.auxiliaryfunctions.write_config(path_config_file,cfg)
print("EXTRACTING FRAMES")
deeplabcut.extract_frames(path_config_file,mode='automatic',userfeedback=False)
print("CREATING-SOME LABELS FOR THE FRAMES")
frames=os.listdir(os.path.join(cfg['project_path'],'labeled-data',videoname))
#As this next step is manual, we update the labels by putting them on the diagonal (fixed for all frames)
for index,bodypart in enumerate(cfg['bodyparts']):
columnindex = pd.MultiIndex.from_product([[scorer], [bodypart], ['x', 'y']],names=['scorer', 'bodyparts', 'coords'])
frame = pd.DataFrame(100+np.ones((len(frames),2))*50*index, columns = columnindex, index = [os.path.join('labeled-data',videoname,fn) for fn in frames])
if index==0:
dataFrame=frame
else:
dataFrame = pd.concat([dataFrame, frame],axis=1)
dataFrame.to_csv(os.path.join(cfg['project_path'],'labeled-data',videoname,"CollectedData_" + scorer + ".csv"))
dataFrame.to_hdf(os.path.join(cfg['project_path'],'labeled-data',videoname,"CollectedData_" + scorer + '.h5'),'df_with_missing',format='table', mode='w')
stoptrain=5 #0
keepdeconvweights=True
print("Plot labels...")
deeplabcut.check_labels(path_config_file)
for shuffle,net_type in enumerate(['mobilenet_v2_0.35','resnet_50']): #'mobilenet_v2_1.0']): # 'resnet_50']):
'''
if shuffle==0:
keepdeconvweights=True
else:
keepdeconvweights=False
'''
print("CREATING TRAININGSET", net_type)
if 'resnet_50'==net_type: #this tests the default condition...
deeplabcut.create_training_dataset(path_config_file,Shuffles=[shuffle],augmenter_type=augmenter_type)
else:
deeplabcut.create_training_dataset(path_config_file,Shuffles=[shuffle],net_type=net_type,augmenter_type=augmenter_type)
Cuttrainingschedule(path_config_file,shuffle,lastvalue=stoptrain)
print("TRAIN")
deeplabcut.train_network(path_config_file,shuffle=shuffle)
print("EVALUATE")
deeplabcut.evaluate_network(path_config_file,Shuffles=[shuffle],plotting=True)
print("CUT SHORT VIDEO AND ANALYZE")
if shuffle==0:
# Make super short video (so the analysis is quick!)
newvideo=deeplabcut.ShortenVideo(video[0],start='00:00:00',stop='00:00:00.4',outsuffix='short',outpath=os.path.join(cfg['project_path'],'videos'))
vname=Path(newvideo).stem
deeplabcut.analyze_videos(path_config_file, [newvideo], shuffle=shuffle, save_as_csv=True, destfolder=dfolder)
print("CREATE VIDEO")
deeplabcut.create_labeled_video(path_config_file,[newvideo],shuffle=shuffle, destfolder=dfolder)
print("Making plots")
deeplabcut.plot_trajectories(path_config_file,[newvideo],shuffle=shuffle, destfolder=dfolder)
print("EXTRACT OUTLIERS")
deeplabcut.extract_outlier_frames(path_config_file,[newvideo],shuffle=shuffle,outlieralgorithm='jump',epsilon=0,automatic=True, destfolder=dfolder)
file=os.path.join(cfg['project_path'],'labeled-data',vname,"machinelabels-iter"+ str(cfg['iteration']) + '.h5')
print("RELABELING")
DF=pd.read_hdf(file,'df_with_missing')
DLCscorer=np.unique(DF.columns.get_level_values(0))[0]
DF.columns.set_levels([scorer.replace(DLCscorer,scorer)],level=0,inplace=True)
DF =DF.drop('likelihood',axis=1,level=2)
DF.to_csv(os.path.join(cfg['project_path'],'labeled-data',vname,"CollectedData_" + scorer + ".csv"))
DF.to_hdf(os.path.join(cfg['project_path'],'labeled-data',vname,"CollectedData_" + scorer + '.h5'),'df_with_missing',format='table', mode='w')
print("MERGING")
deeplabcut.merge_datasets(path_config_file)
print("CREATING TRAININGSET")
deeplabcut.create_training_dataset(path_config_file,Shuffles=[shuffle],net_type=net_type)
Cuttrainingschedule(path_config_file,shuffle,lastvalue=stoptrain,initweights='previteration')
print("TRAINING from previous snapshot!!!!!")
deeplabcut.train_network(path_config_file,shuffle=shuffle,keepdeconvweights=keepdeconvweights)
print("ANALYZING some individual frames")
deeplabcut.analyze_time_lapse_frames(path_config_file,os.path.join(cfg['project_path'],'labeled-data/reachingvideo1/'),shuffle=shuffle)
print("ALL DONE!!! - Mobilnets and ResNets are good!")