| from sklearn.metrics import mean_squared_error, mean_absolute_error
|
| from scipy.io import loadmat
|
| import os
|
| import re
|
|
|
| root_path = "."
|
|
|
|
|
| def get_seq_class(seq, set):
|
| backlight = [
|
| "DJI_0021",
|
| "DJI_0022",
|
| "DJI_0032",
|
| "DJI_0202",
|
| "DJI_0339",
|
| "DJI_0340",
|
| "DJI_0463",
|
| "DJI_0003",
|
| ]
|
|
|
| fly = [
|
| "DJI_0177",
|
| "DJI_0174",
|
| "DJI_0022",
|
| "DJI_0180",
|
| "DJI_0181",
|
| "DJI_0200",
|
| "DJI_0544",
|
| "DJI_0012",
|
| "DJI_0178",
|
| "DJI_0343",
|
| "DJI_0185",
|
| "DJI_0195",
|
| "DJI_0996",
|
| "DJI_0977",
|
| "DJI_0945",
|
| "DJI_0946",
|
| "DJI_0091",
|
| "DJI_0442",
|
| "DJI_0466",
|
| "DJI_0459",
|
| "DJI_0464",
|
| ]
|
|
|
| angle_90 = [
|
| "DJI_0179",
|
| "DJI_0186",
|
| "DJI_0189",
|
| "DJI_0191",
|
| "DJI_0196",
|
| "DJI_0190",
|
| "DJI_0070",
|
| "DJI_0091",
|
| ]
|
|
|
| mid_size = [
|
| "DJI_0012",
|
| "DJI_0013",
|
| "DJI_0014",
|
| "DJI_0021",
|
| "DJI_0022",
|
| "DJI_0026",
|
| "DJI_0028",
|
| "DJI_0028",
|
| "DJI_0030",
|
| "DJI_0028",
|
| "DJI_0030",
|
| "DJI_0034",
|
| "DJI_0200",
|
| "DJI_0544",
|
| "DJI_0463",
|
| "DJI_0001",
|
| "DJI_0149",
|
| ]
|
|
|
| light = "sunny"
|
| bird = "stand"
|
| angle = "60"
|
| size = "small"
|
|
|
| if seq in backlight:
|
| light = "backlight"
|
| if seq in fly:
|
| bird = "fly"
|
| if seq in angle_90:
|
| angle = "90"
|
| if seq in mid_size:
|
| size = "mid"
|
|
|
| count = "sparse"
|
| loca = loadmat(
|
| os.path.join(
|
| root_path,
|
| set,
|
| "ground_truth",
|
| "GT_img" + str(seq[-3:]) + "000.mat",
|
| )
|
| )["locations"]
|
| if loca.shape[0] > 150:
|
| count = "crowded"
|
| return [light, angle, bird, size, count]
|
|
|
|
|
| with open("./result.txt", "r") as f:
|
| lines = f.readlines()
|
| print(len(lines))
|
|
|
| data = []
|
| for line in lines:
|
| match = re.match(
|
| r"\d+: *err: *(\d+), *gt_count: *(\d+), *pred_count: *(\d+), *name: *(.+)",
|
|
|
|
|
| line,
|
| )
|
| if match:
|
| error = float(match.group(1))
|
| gt_count = float(match.group(2))
|
| pred_count = float(match.group(3))
|
| name = match.group(4)
|
| data.append(
|
| {
|
| "error": error,
|
| "gt_count": gt_count,
|
| "pred_count": pred_count,
|
| "name": name,
|
| }
|
| )
|
| print(error, gt_count, pred_count, name)
|
|
|
| print(len(data))
|
|
|
|
|
| preds = []
|
| gts = []
|
| preds_hist = [[] for i in range(10)]
|
| gts_hist = [[] for i in range(10)]
|
| attri = [
|
| "sunny",
|
| "backlight",
|
| "crowded",
|
| "sparse",
|
| "60",
|
| "90",
|
| "stand",
|
| "fly",
|
| "small",
|
| "mid",
|
| ]
|
|
|
| for d in data:
|
| name = d["name"]
|
| error = d["error"]
|
| gt_count = d["gt_count"]
|
| pred_count = d["pred_count"]
|
| seq = "DJI_" + str(int(name[3:6])).zfill(4)
|
| cur_attris = get_seq_class(seq, "test")
|
| preds.append(pred_count)
|
| gts.append(gt_count)
|
| for cur_attri in cur_attris:
|
| preds_hist[attri.index(cur_attri)].append(pred_count)
|
| gts_hist[attri.index(cur_attri)].append(gt_count)
|
| test_log = "Test: MAE: {:.2f}, MSE: {:.2f}\n".format(
|
| mean_absolute_error(gts, preds), mean_squared_error(gts, preds)
|
| )
|
| for i in range(10):
|
| if len(preds_hist[i]) == 0:
|
| continue
|
| test_log_attri = "{}: MAE: {:.2f}, MSE: {:.2f}\n".format(
|
| attri[i],
|
| mean_absolute_error(gts_hist[i], preds_hist[i]),
|
| mean_squared_error(gts_hist[i], preds_hist[i]),
|
| )
|
| test_log += test_log_attri
|
| print(test_log)
|
|
|
