DeepLabCut 개요 번역

import deeplabcut
deeplabcut.create_new_project('ProjectName','YourName', ['/usr/FullPath/OfVideo1.avi', '/usr/FullPath/OfVideo2.avi', '/usr/FullPath/OfVideo1.avi'],
              copy_videos=True/False, multianimal=True/False)

import os
os.environ['DLClight'] = "True"
os.environ["CUDA_DEVICE_ORDER"]="PCI_BUS_ID"   
os.environ["CUDA_VISIBLE_DEVICES"]="0,1"
import deeplabcut

import pandas as pd
import os
import numpy as np
import h5py

####### 고정적인 값들 ########
scorer = 'eastRain'
origin_file = 'labeled-data/origin/CollectedData_' + scorer + '_origin.h5'

####### 학습 시킬 때 변경해야 하는 변수들 #########
iteration_num = 3
train_data = '/home/eastrain/mnt/test_custom_data/0519'

####### 자동으로 완성되는 변수들(딱히 건드릴일은 없다) ##########
iteration_name = 'iteration-' + str(iteration_num)
output_file = os.path.join('labeled-data', iteration_name,'CollectedData_' + scorer + '.h5')
config_path = 'config_' + str(iteration_num) + '.yaml'

bodyparts = [
    'eye-left',
    'eye-right',
    'nose',
    'mouth-left',
    'mouth-right',
    'mouth-low',
    'tongue',
    'ear-left-start',
    'ear-left-middle',
    'ear-left-end',
    'ear-right-start',
    'ear-right-middle',
    'ear-right-end',
    'withers',
    'prosternum',
    'elbow-left',
    'forearm-left',
    'forefoot-left',
    'elbow-right',
    'forearm-right',
    'forefoot-right',
    'stiffle-left',
    'hock-left',
    'hindfoot-left',
    'stiffle-right',
    'hock-right',
    'hindfoot-right',
    'tail-start',
    'tail-start-middle',
    'tail-middle',
    'tail-middle-end',
    'tail-end'
]

nan = np.empty((1,2,))
nan[:] = np.nan
dataFrame = pd.read_hdf(origin_file,'df_with_missing')
cnt = 0
all_csv_files = []

for (path, directory, files) in os.walk(train_data):
    for filename in files:
        file_abs_path = os.path.join(path, filename)
        ext = str(os.path.splitext(file_abs_path)[1]).lower()
        if ext == '.csv':
            all_csv_files.append(str(file_abs_path))

total_cnt = len(all_csv_files)

for file in all_csv_files:
    csv = pd.read_csv(file, header=None, usecols=[0,1,2,3,4], names=['key', '1', '2', '3', '4'])
    file_name = file.split('.')[0] + '.' + csv['key'][0].split('.')[1]
    temp_df=None
    
    point_dict = {}
    for _index, item in enumerate(csv['key'][3:]):
        if item == 'eye-left-open':
            item = 'eye-left'
        elif item == 'eye-right-open':
            item = 'eye-right'        
        
        point_dict[item] = [csv['1'][3 + _index], csv['2'][3 + _index]]
    
    for bodypart in bodyparts:
        index = pd.MultiIndex.from_product([[scorer], [bodypart], ['x', 'y']],names=['scorer', 'bodyparts', 'coords'])
        
        if bodypart in point_dict:
            frame = pd.DataFrame([point_dict[bodypart]], columns = index, index = [file_name])
        else:
            frame = pd.DataFrame(nan, columns = index, index = [file_name])
        temp_df = pd.concat([temp_df, frame],axis=1)
    dataFrame = pd.concat([dataFrame, temp_df],axis=0)
    
    #for _index, item in enumerate(csv['key'][3:]):
    #    if item == 'eye-left-open':
    #        item = 'eye-left'
    #    elif item == 'eye-right-open':
    #        item = 'eye-right'        
    #    
    #    dataFrame.loc[file_name][scorer, item,'x' ] = csv['1'][3 + _index]
    #    dataFrame.loc[file_name][scorer, item,'y' ] = csv['2'][3 + _index]
        
    cnt += 1
    print(str(cnt) + " / " + str(total_cnt))


labeled_dir = os.path.join('labeled-data', iteration_name)
if not os.path.isdir(labeled_dir):
    os.mkdir(labeled_dir)
labeled_file_root = os.path.join(labeled_dir, 'CollectedData_' + scorer)
dataFrame.to_csv(labeled_file_root + '.csv')
dataFrame.to_hdf(labeled_file_root + '.h5', 'df_with_missing',format='table', mode='w')

deeplabcut.create_training_dataset(config_path)

deeplabcut.train_network(config_path, shuffle=1, displayiters=1000, saveiters=50000, max_snapshots_to_keep=10, maxiters=1330000, gputouse=0,1)

import deeplabcut

config_path = 'config.yaml'
deeplabcut.evaluate_network(config_path, plotting=False, gputouse=0,1, rescale=True)

file_name = 'dot_eatin'
file_type = 'mp4'
iteration = '3'
config_path = '/home/eastrain/workspace/0514/config_' + iteration +'.yaml'
target = os.path.join('/home/eastrain/workspace/0514/output/iteration-' + iteration, file_name, file_name + '.' + file_type)

deeplabcut.analyze_videos(config_path,[target], videotype='.mp4', save_as_csv = True, gputouse=0)

deeplabcut.filterpredictions(config_path,[target])

deeplabcut.analyzeskeleton(config_path, [target], videotype='.mp4', shuffle=1, trainingsetindex=0, save_as_csv=True, destfolder=None)

deeplabcut.create_labeled_video(config_path, [target], filtered=True, draw_skeleton = True)

deeplabcut.plot_trajectories(config_path,[target],filtered = True, showfigures= True)