Upload DroneBird/toolkit/eval.py
Browse files- DroneBird/toolkit/eval.py +175 -0
DroneBird/toolkit/eval.py
ADDED
|
@@ -0,0 +1,175 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from sklearn.metrics import mean_squared_error, mean_absolute_error
|
| 2 |
+
from scipy.io import loadmat
|
| 3 |
+
import os
|
| 4 |
+
import re
|
| 5 |
+
|
| 6 |
+
root_path = "."
|
| 7 |
+
|
| 8 |
+
|
| 9 |
+
def get_seq_class(seq, set):
|
| 10 |
+
backlight = [
|
| 11 |
+
"DJI_0021",
|
| 12 |
+
"DJI_0022",
|
| 13 |
+
"DJI_0032",
|
| 14 |
+
"DJI_0202",
|
| 15 |
+
"DJI_0339",
|
| 16 |
+
"DJI_0340",
|
| 17 |
+
"DJI_0463",
|
| 18 |
+
"DJI_0003",
|
| 19 |
+
]
|
| 20 |
+
|
| 21 |
+
fly = [
|
| 22 |
+
"DJI_0177",
|
| 23 |
+
"DJI_0174",
|
| 24 |
+
"DJI_0022",
|
| 25 |
+
"DJI_0180",
|
| 26 |
+
"DJI_0181",
|
| 27 |
+
"DJI_0200",
|
| 28 |
+
"DJI_0544",
|
| 29 |
+
"DJI_0012",
|
| 30 |
+
"DJI_0178",
|
| 31 |
+
"DJI_0343",
|
| 32 |
+
"DJI_0185",
|
| 33 |
+
"DJI_0195",
|
| 34 |
+
"DJI_0996",
|
| 35 |
+
"DJI_0977",
|
| 36 |
+
"DJI_0945",
|
| 37 |
+
"DJI_0946",
|
| 38 |
+
"DJI_0091",
|
| 39 |
+
"DJI_0442",
|
| 40 |
+
"DJI_0466",
|
| 41 |
+
"DJI_0459",
|
| 42 |
+
"DJI_0464",
|
| 43 |
+
]
|
| 44 |
+
|
| 45 |
+
angle_90 = [
|
| 46 |
+
"DJI_0179",
|
| 47 |
+
"DJI_0186",
|
| 48 |
+
"DJI_0189",
|
| 49 |
+
"DJI_0191",
|
| 50 |
+
"DJI_0196",
|
| 51 |
+
"DJI_0190",
|
| 52 |
+
"DJI_0070",
|
| 53 |
+
"DJI_0091",
|
| 54 |
+
]
|
| 55 |
+
|
| 56 |
+
mid_size = [
|
| 57 |
+
"DJI_0012",
|
| 58 |
+
"DJI_0013",
|
| 59 |
+
"DJI_0014",
|
| 60 |
+
"DJI_0021",
|
| 61 |
+
"DJI_0022",
|
| 62 |
+
"DJI_0026",
|
| 63 |
+
"DJI_0028",
|
| 64 |
+
"DJI_0028",
|
| 65 |
+
"DJI_0030",
|
| 66 |
+
"DJI_0028",
|
| 67 |
+
"DJI_0030",
|
| 68 |
+
"DJI_0034",
|
| 69 |
+
"DJI_0200",
|
| 70 |
+
"DJI_0544",
|
| 71 |
+
"DJI_0463",
|
| 72 |
+
"DJI_0001",
|
| 73 |
+
"DJI_0149",
|
| 74 |
+
]
|
| 75 |
+
|
| 76 |
+
light = "sunny"
|
| 77 |
+
bird = "stand"
|
| 78 |
+
angle = "60"
|
| 79 |
+
size = "small"
|
| 80 |
+
# resolution = '4k'
|
| 81 |
+
if seq in backlight:
|
| 82 |
+
light = "backlight"
|
| 83 |
+
if seq in fly:
|
| 84 |
+
bird = "fly"
|
| 85 |
+
if seq in angle_90:
|
| 86 |
+
angle = "90"
|
| 87 |
+
if seq in mid_size:
|
| 88 |
+
size = "mid"
|
| 89 |
+
|
| 90 |
+
count = "sparse"
|
| 91 |
+
loca = loadmat(
|
| 92 |
+
os.path.join(
|
| 93 |
+
root_path,
|
| 94 |
+
set,
|
| 95 |
+
"ground_truth",
|
| 96 |
+
"GT_img" + str(seq[-3:]) + "000.mat",
|
| 97 |
+
)
|
| 98 |
+
)["locations"]
|
| 99 |
+
if loca.shape[0] > 150:
|
| 100 |
+
count = "crowded"
|
| 101 |
+
return [light, angle, bird, size, count]
|
| 102 |
+
|
| 103 |
+
|
| 104 |
+
with open("./result.txt", "r") as f:
|
| 105 |
+
lines = f.readlines()
|
| 106 |
+
print(len(lines))
|
| 107 |
+
|
| 108 |
+
data = []
|
| 109 |
+
for line in lines:
|
| 110 |
+
match = re.match(
|
| 111 |
+
r"\d+: *err: *(\d+), *gt_count: *(\d+), *pred_count: *(\d+), *name: *(.+)",
|
| 112 |
+
# r'\d+: err: (\d+\.\d+), gt_count: (\d+\.\d+), pred_count: (\d+\.\d+), name: (.+)',
|
| 113 |
+
# r'vi: \d+, name: (.+), gt: (\d+\.\d+), pred: (\d+\.\d+), res: (\d+\.\d+)',
|
| 114 |
+
line,
|
| 115 |
+
)
|
| 116 |
+
if match:
|
| 117 |
+
error = float(match.group(1))
|
| 118 |
+
gt_count = float(match.group(2))
|
| 119 |
+
pred_count = float(match.group(3))
|
| 120 |
+
name = match.group(4)
|
| 121 |
+
data.append(
|
| 122 |
+
{
|
| 123 |
+
"error": error,
|
| 124 |
+
"gt_count": gt_count,
|
| 125 |
+
"pred_count": pred_count,
|
| 126 |
+
"name": name,
|
| 127 |
+
}
|
| 128 |
+
)
|
| 129 |
+
print(error, gt_count, pred_count, name)
|
| 130 |
+
|
| 131 |
+
print(len(data))
|
| 132 |
+
|
| 133 |
+
|
| 134 |
+
preds = []
|
| 135 |
+
gts = []
|
| 136 |
+
preds_hist = [[] for i in range(10)]
|
| 137 |
+
gts_hist = [[] for i in range(10)]
|
| 138 |
+
attri = [
|
| 139 |
+
"sunny",
|
| 140 |
+
"backlight",
|
| 141 |
+
"crowded",
|
| 142 |
+
"sparse",
|
| 143 |
+
"60",
|
| 144 |
+
"90",
|
| 145 |
+
"stand",
|
| 146 |
+
"fly",
|
| 147 |
+
"small",
|
| 148 |
+
"mid",
|
| 149 |
+
]
|
| 150 |
+
|
| 151 |
+
for d in data:
|
| 152 |
+
name = d["name"]
|
| 153 |
+
error = d["error"]
|
| 154 |
+
gt_count = d["gt_count"]
|
| 155 |
+
pred_count = d["pred_count"]
|
| 156 |
+
seq = "DJI_" + str(int(name[3:6])).zfill(4)
|
| 157 |
+
cur_attris = get_seq_class(seq, "test")
|
| 158 |
+
preds.append(pred_count)
|
| 159 |
+
gts.append(gt_count)
|
| 160 |
+
for cur_attri in cur_attris:
|
| 161 |
+
preds_hist[attri.index(cur_attri)].append(pred_count)
|
| 162 |
+
gts_hist[attri.index(cur_attri)].append(gt_count)
|
| 163 |
+
test_log = "Test: MAE: {:.2f}, MSE: {:.2f}\n".format(
|
| 164 |
+
mean_absolute_error(gts, preds), mean_squared_error(gts, preds)
|
| 165 |
+
)
|
| 166 |
+
for i in range(10):
|
| 167 |
+
if len(preds_hist[i]) == 0:
|
| 168 |
+
continue
|
| 169 |
+
test_log_attri = "{}: MAE: {:.2f}, MSE: {:.2f}\n".format(
|
| 170 |
+
attri[i],
|
| 171 |
+
mean_absolute_error(gts_hist[i], preds_hist[i]),
|
| 172 |
+
mean_squared_error(gts_hist[i], preds_hist[i]),
|
| 173 |
+
)
|
| 174 |
+
test_log += test_log_attri
|
| 175 |
+
print(test_log)
|