def _get_ecp_day_night_9_priors():
priors = [[0.5728529907421875, 0.13943622409895834],
[0.41761617583007815, 0.09156660707291667],
[0.3015263176855469, 0.06248444700520834],
[0.22101856140625, 0.042888710765625],
[0.1533158565527344, 0.031196821406250002],
[0.11255495265625, 0.021566710822916668],
[0.07823327209960937, 0.015212825187500001],
[0.0533416983203125, 0.010216603067708333],
[0.0332035418359375, 0.006413999807291667]]
# # in pixels:
# priors = [[586.60146252, 267.71755027],
# [427.63896405, 175.80788558],
# [308.76294931, 119.97013825],
# [226.32300688, 82.34632467],
# [156.99543711, 59.8978971],
# [115.25627152, 41.40808478],
# [80.11087063, 29.20862436],
# [54.62189908, 19.61587789],
# [34.00042684, 12.31487963]]
priors_32 = [data.Prior(h=p[0], w=p[1]) for p in priors[:3]]
priors_16 = [data.Prior(h=p[0], w=p[1]) for p in priors[3:6]]
priors_8 = [data.Prior(h=p[0], w=p[1]) for p in priors[6:]]
return {
32: priors_32,
16: priors_16,
8: priors_8,
}
def _get_ecp_night_9_priors():
priors = [
[0.6197282176953125, 0.14694562146874998],
[0.4243941425683594, 0.09687759120833334],
[0.3103862368359375, 0.06362734035416667],
[0.23494613041992188, 0.043568554453125],
[0.1634832566796875, 0.03293052755208333],
[0.12444031231445313, 0.023274527578125],
[0.08800429220703125, 0.016930080526041665],
[0.06101826478515625, 0.011638404229166668],
[0.03925641140625, 0.007475639645833334],
]
# # in pixels:
# priors = [[634.60169492, 282.13559322],
# [434.57960199, 186.00497512],
# [317.83550652, 122.16449348],
# [240.58483755, 83.65162455],
# [167.40685484, 63.2266129],
# [127.42687981, 44.68709295],
# [90.11639522, 32.50575461],
# [62.48270314, 22.34573612],
# [40.19856528, 14.35322812]]
priors_32 = [data.Prior(h=p[0], w=p[1]) for p in priors[:3]]
priors_16 = [data.Prior(h=p[0], w=p[1]) for p in priors[3:6]]
priors_8 = [data.Prior(h=p[0], w=p[1]) for p in priors[6:]]
return {
32: priors_32,
16: priors_16,
8: priors_8,
}
def _get_ecp_9_priors():
priors = [
[0.56643243, 0.13731691],
[0.41022839, 0.09028599],
[0.30508716, 0.06047965],
[0.20774711, 0.04376083],
[0.15475611, 0.02996197],
[0.10878717, 0.02149197],
[0.07694039, 0.01488527],
[0.05248527, 0.01007212],
[0.03272104, 0.00631827],
]
# # in pixels:
# priors = [[580.02680832, 263.64846719999997],
# [420.07387136, 173.3491008],
# [312.40925184, 116.120928],
# [212.73304064, 84.0207936],
# [158.47025664, 57.5269824],
# [111.39806208, 41.264582399999995],
# [78.78695936, 28.5797184],
# [53.74491648, 19.338470400000002],
# [33.50634496, 12.1310784]]
priors_32 = [data.Prior(h=p[0], w=p[1]) for p in priors[:3]]
priors_16 = [data.Prior(h=p[0], w=p[1]) for p in priors[3:6]]
priors_8 = [data.Prior(h=p[0], w=p[1]) for p in priors[6:]]
return {
32: priors_32,
16: priors_16,
8: priors_8,
}
def _get_ecp_with_bic_9_priors():
priors = [
[0.5541169062011718, 0.15767184942708334],
[0.3872792363671875, 0.08849276056770834],
[0.27297898112304686, 0.05552458755208333],
[0.18570756796875, 0.034849724458333335],
[0.13080457012695312, 0.052510955223958336],
[0.12203939466796875, 0.02422101765625],
[0.083340965234375, 0.01635016602083333],
[0.055563667021484374, 0.010672233619791667],
[0.03409191838867188, 0.006481136984375],
]
# # in pixels:
# priors = [[567.41571195, 302.7299509],
# [396.57393804, 169.90610029],
# [279.53047667, 106.6072081],
# [190.1645496, 66.91147096],
# [133.94387981, 100.82103403],
# [124.96834014, 46.5043539],
# [85.3411484, 31.39231876],
# [56.89719503, 20.49068855],
# [34.91012443, 12.44378301]]
priors_32 = [data.Prior(h=p[0], w=p[1]) for p in priors[:3]]
priors_16 = [data.Prior(h=p[0], w=p[1]) for p in priors[3:6]]
priors_8 = [data.Prior(h=p[0], w=p[1]) for p in priors[6:]]
return {
32: priors_32,
16: priors_16,
8: priors_8,
}
def _get_city_persons_9_priors():
priors = [[495.27, 203.83], [297.84, 122.19], [197.44, 81.48],
[141.07, 58.5], [102.72, 43.1], [75.78, 31.66], [54.24, 23.19],
[37.55, 16.15], [22.55, 10.09]]
priors = [[p[0] / 1024., p[1] / 2048.] for p in priors
] # priors are calculated for original citypersons img size
priors_32 = [data.Prior(h=p[0], w=p[1]) for p in priors[:3]]
priors_16 = [data.Prior(h=p[0], w=p[1]) for p in priors[3:6]]
priors_8 = [data.Prior(h=p[0], w=p[1]) for p in priors[6:]]
return {
32: priors_32,
16: priors_16,
8: priors_8,
}
def img_size_and_priors_if_crop(
config): # TODO fn name is not very descriptive
img_size = config['crop_img_size'] if config['crop'] else config[
'full_img_size']
priors = config['priors']
if config['crop']:
# priors are always defined for the full img => need to rescale if we crop the image
scale_h = config['full_img_size'][0] / float(
config['crop_img_size'][0])
scale_w = config['full_img_size'][1] / float(
config['crop_img_size'][1])
for stride, prs in priors.items():
priors[stride] = [
data.Prior(h=p.h * scale_h, w=p.w * scale_w) for p in prs
]
return img_size, priors