Dipeshpal

# <Filter>: 
#      None(default): load all. each has a bbox. some instances have keypoint and some have mask annotations.
#            images: 5081, instances: 13360
#     'kpt&segm' or 'segm&kpt': only load instances contained both keypoint and mask annotations (and bbox)
#            images: 4731, instances: 8110
#     'kpt|segm' or 'segm|kpt': load instances contained either keypoint or mask annotations (and bbox)
#            images: 5081, instances: 10375
#     'kpt' or 'segm': load instances contained particular kind of annotations (and bbox)
#            images: 5081/4731, instances: 10375/8110
ochuman = OCHuman(AnnoFile='ochuman.json', Filter='segm')

Dipeshpal

def get_segmentation(data):
    img = cv2.imread(os.path.join(ImgDir, data['file_name']))
    height, width = data['height'], data['width']

    colors = [[255, 0, 0], 
            [255, 255, 0],
            [0, 255, 0],
            [0, 255, 255], 
            [0, 0, 255], 
            [255, 0, 255]]

Dipeshpal

def new_mask(real_img, m_img):
    real_img = real_img.reshape(1, -1)[0]
    m_img = m_img.reshape(1, -1)[0]

    new = []

    for i, j in zip(real_img, m_img):
        if i != j:
            new.append(255) # human will white appear because of 255
        else:

Dipeshpal

IMG_HEIGHT = 512
IMG_WIDTH = 512
ImgDir = 'images/'

Dipeshpal

from ochumanApi.ochuman import OCHuman
import cv2, os
import matplotlib
import matplotlib.pyplot as plt
%matplotlib inline
plt.rcParams['figure.figsize'] = (15, 15)

import ochumanApi.vis as vistool
from ochumanApi.ochuman import Poly2Mask

Dipeshpal

def generator_images(batch_size=1, ind):
    while True:
        x_batch = []
        y_batch = []

        for i in range(batch_size):
            data = ochuman.loadImgs(imgIds=[image_ids[ind]])[0]

            file_name = data['file_name']

Dipeshpal

for i in tqdm(range(4731)):
    for x, y in generator_images(1, i):
        break

    base_dir_custom = "custom_dataset_human_black_background/"
    try:
        os.makedirs(f'{base_dir_custom}')
    except:
        pass
    try:

Dipeshpal

Model: "model"
__________________________________________________________________________________________________
Layer (type)                    Output Shape         Param #     Connected to                     
==================================================================================================
input_1 (InputLayer)            [(None, 512, 512, 3) 0                                            
__________________________________________________________________________________________________
conv2d (Conv2D)                 (None, 512, 512, 32) 896         input_1[0][0]                    
__________________________________________________________________________________________________
dropout (Dropout)               (None, 512, 512, 32) 0           conv2d[0][0]                     
__________________________________________________________________________________________________

Dipeshpal

model_name = "models/"+"Unet_black_background.h5"

modelcheckpoint = ModelCheckpoint(model_name,
                                  monitor='val_loss',
                                  mode='auto',
                                  verbose=1,
                                  save_best_only=True)

lr_callback = ReduceLROnPlateau(min_lr=0.000001)

Dipeshpal

Training: epoch {} begins at 18:08:48.853186
Epoch 1/50
100/100 [==============================] - ETA: 0s - loss: 0.5708 - acc: 0.2714 
Epoch 00001: val_loss improved from inf to 0.52510, saving model to models/Unet_black_background.h5
Training: epoch 0 ends at 19:03:21.708891
100/100 [==============================] - 3295s 33s/step - loss: 0.5708 - acc: 0.2714 - val_loss: 0.5251 - val_acc: 0.1151 - lr: 0.0010
Epoch 2/50
100/100 [==============================] - ETA: 0s - loss: 0.5104 - acc: 0.2918 
Epoch 00002: val_loss improved from 0.52510 to 0.50866, saving model to models/Unet_black_background.h5
Training: epoch 1 ends at 19:27:18.558237