Training

train.py

import numpy as np
from pycocotools.coco import COCO
import os
import math
import keras
from keras.models import Sequential, Model
from keras.layers import Dense, Activation, Input, Flatten, Dropout
from keras.utils import plot_model
import cv2
import matplotlib.pyplot as plt

def bbox_generator():
    weights = np.zeros((90, 140, 17))
    output = np.zeros((90, 140, 17))
    coco = COCO(os.path.join('', 'annotations', 'person_keypoints_val2017.json'))
    thres = 0.05
    ann_ids = coco.getAnnIds()

    while 1:
        #for i in range(1875):
        #    if i%125==0:
        #        print "i = " + str(i)
        #    yield
        print "listen up"

        for i in range(len(ann_ids)):
            print i
            ann_data = coco.loadAnns(2203726)[0]
            # ann_ids[i]
            bbox = ann_data['bbox']
            x0 = int(bbox[0])
            y0 = int(bbox[1])
            x = float(bbox[2])
            y = float(bbox[3])
            xscale = 90.0 / math.ceil(x)
            yscale = 140.0 / math.ceil(y)

            kpx = ann_data['keypoints'][0::3]
            kpy = ann_data['keypoints'][1::3]
            kpv = ann_data['keypoints'][2::3]

            for xy in range(17):
                if kpv[xy] > 0:
                    a = int(round(kpx[xy] - x0) * xscale)
                    b = int(round(kpy[xy] - y0) * yscale)
                    if a >= 90 and b >= 140:
                        weights[89][139][xy] = 1
                        continue
                    if a >= 90:
                        weights[89][b][xy] = 1
                    if b >= 140:
                        weights[a][139][xy] = 1
                    else:
                        output[a, b, xy] = 1

            img_id = ann_data['image_id']
            img_data = coco.loadImgs(img_id)[0]
            ann_data = coco.loadAnns(coco.getAnnIds(img_data['id']))
            for a in range(len(ann_data)):
                kpx = ann_data[a]['keypoints'][0::3]
                kpy = ann_data[a]['keypoints'][1::3]
                kpv = ann_data[a]['keypoints'][2::3]
                for x in range(17):
                    if kpv[x] > 0:

                        if (kpx[x] > bbox[0] - bbox[2] * thres and kpx[x] < bbox[0] + bbox[2] * (1 + thres)):
                            if (kpy[x] > bbox[1] - bbox[3] * thres and kpy[x] < bbox[1] + bbox[3] * (1 + thres)):
                                a = int(round(kpx[x] - x0) * xscale)
                                b = int(round(kpy[x] - y0) * yscale)
                                #print(ann_data[i]['image_id'])
                                if a >= 90 and b >= 140:
                                    weights[89][139][x] = 1
                                    print(x0, xscale, bbox[2], a, b)
                                    continue
                                if a >= 90:
                                    weights[89][b][x] = 1
                                if b >= 140:
                                    weights[a][139][x] = 1
                                if a < 90 and b < 140:
                                    weights[a][b][x] = 1
            yield np.expand_dims(weights,axis=0), np.expand_dims(output,axis=0)

#model = Sequential()

#inputs = Input(shape=(214200))

visible = Input(shape=(90,140,17))
#flatten = Flatten()(visible)
hidden1 = Dense(17, activation='relu')(visible)
#flatten2 = Flatten()(hidden1)
#output = Dense(107100, activation='relu')(flatten2)
model = Model(inputs=visible, outputs=hidden1)

# Summarize Layers
print(model.summary())

# Plot Graph
plot_model(model, to_file='test.png')

model.compile(loss='categorical_crossentropy', optimizer='sgd')

model.fit_generator(bbox_generator(), samples_per_epoch = 6000, nb_epoch=10, verbose=2,callbacks=[], validation_data=None)