COLO/export_results.py at main · Niche-Lab/COLO · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
import pandas as pd
import numpy as np
from tensorboard.backend.event_processing import event_accumulator

from path import PathFinder
PATHS = PathFinder()
DIR_OUT = PATHS["DIR_SRC"] / "out"
DICT_PARAMS = dict({
    "rtdetr-l": 45,
    "rtdetr-x": 86,
    "yolo12n": 2.6,
    "yolo12m": 20.2,
    "yolo12x": 59.1,
    "yolo11n": 2.6,
    "yolo11m": 20.1,
    "yolo11x": 56.9,
})

def main():
    ls_csv = [f for f in (DIR_OUT / "study1").glob("*.csv")] + \
             [f for f in (DIR_OUT / "study2").glob("*.csv")]

    # study 1
    df_s1 = pd.concat([pd.read_csv(f) for f in ls_csv if "study1" in str(f)], ignore_index=True).drop_duplicates()
    df_s1 = df_s1.query("n in [32, 128, 500]")
    df_s1 = add_family(df_s1)
    df_s1 = add_size(df_s1)
    df_s1 = df_s1.query("iter <= 2")

    # study 2
    df_s2 = pd.concat([pd.read_csv(f) for f in ls_csv if "study2" in str(f)], ignore_index=True).drop_duplicates()
    df_s2 = add_family(df_s2)
    df_s2 = add_size(df_s2)

    # study time
    tensorboard_files = list(DIR_OUT.rglob("events.out.tfevents.*"))
    df_time = get_med_time_from_tensorboard(tensorboard_files)
    df_time.head()
    df_time = add_family(df_time)
    df_time = add_size(df_time)
    df_time = add_gpu(df_time)
    df_time = add_params(df_time)

    # export
    df_s1.to_csv(DIR_OUT / "study1.csv", index=False)
    df_s2.to_csv(DIR_OUT / "study2.csv", index=False)
    df_time.to_csv(DIR_OUT / "train_time.csv", index=False)

def add_family(df):
    df.loc[:, ["family"]] = "DETR"
    df.loc[df["model"].str.contains("yolo12"), "family"] = "YOLOv12"
    df.loc[df["model"].str.contains("yolo11"), "family"] = "YOLOv11"
    return df

def add_size(df):
    df.loc[:, ["size"]] = "small"
    df.loc[df["model"].str.contains("yolo11x"), "size"] = "large"
    df.loc[df["model"].str.contains("yolo12x"), "size"] = "large"
    df.loc[df["model"].str.contains("rtdetr-x"), "size"] = "large"
    return df

def add_gpu(df):
    df.loc[:, "gpu"] = ["a100 (80GB)" if t < 5 else "l40s (48GB)" for t in df["thread"]]
    return df

def add_params(df):
    df.loc[:, "params"] = df["model"].map(DICT_PARAMS)
    return df

def get_med_time_from_tensorboard(tensorboard_files):
    ls_select = []
    ls_models = []
    ls_thread = []
    ls_mu = []
    for m in ["yolo12x", "yolo11x", "rtdetr-x", "rtdetr-l", "yolo12n", "yolo11n"]:
        for t in range(10):
            for i in range(2):
                filename = f"study1/thread_{t}/0_all_{m}_128/iter_{i}/events.out"
                match_file = [f for f in tensorboard_files if filename in str(f)]
                if len(match_file) == 0:
                    continue
                path_file = PATHS["DIR_SRC"] / "out" / match_file[0]
                mu = get_med_time(path_file)
                ls_mu.append(mu)
                ls_models.append(m)
                ls_thread.append(t)
                ls_select.append(filename)
    data = pd.DataFrame({
        "model": ls_models,
        "thread": ls_thread,
        "median": ls_mu,
        "filename": ls_select,
    })
    return data

def get_med_time(tfile):
    ea = event_accumulator.EventAccumulator(str(tfile))
    ea.Reload()
    events = ea.Scalars("train/cls_loss")
    step_times = []
    for i in range(1, len(events)):
        delta_time = events[i].wall_time - events[i - 1].wall_time
        delta_step = events[i].step - events[i - 1].step
        if delta_step > 0:
            step_times.append(delta_time / delta_step)
    med = np.median(step_times)
    return med


if __name__ == "__main__":
    main()


# import pandas as pd

# from export_results import add_family, add_params, add_size

# data = pd.read_csv("mem.csv")
# ls_models = ["yolo12x", "yolo11x", "rtdetr-x", "rtdetr-l", "yolo12n", "yolo11n"]
# ls_size = [32, 128, 500]
# data.loc[:, "model"] = [m for m in ls_models for _ in range(12)]
# data.loc[:, "n"] = [s for _ in ls_models for i in range(4) for s in ls_size ]
# data

# import seaborn as sns
# import matplotlib.pyplot as plt

# # y = mem, x = n, color = model
# sns.set(style="whitegrid")
# plt.figure(figsize=(10, 6))
# sns.lineplot(data=data, x="n", y="mem", hue="model", marker="o")


# data = add_family(data)
# data = add_size(data)
# data = add_params(data)

# data.to_csv("train_mem.csv", index=False)