Files
RoboGauge/robogauge/utils/visualize_results.py

171 lines
6.0 KiB
Python

import matplotlib.pyplot as plt
from pathlib import Path
import numpy as np
import yaml
import argparse
import os
# --- 配置 Matplotlib 样式 ---
config = {
"font.family": 'serif',
"figure.figsize": (8, 6),
"font.size": 12,
"mathtext.fontset": 'cm',
'axes.unicode_minus': False
}
plt.rcParams.update(config)
plt.rcParams['font.serif'] = ['Times New Roman', 'DejaVu Serif', 'serif']
def parse_value_string(val_str):
if isinstance(val_str, (int, float)):
return float(val_str)
if isinstance(val_str, str):
if '±' in val_str:
return float(val_str.split('±')[0].strip())
return float(val_str)
return 0.0
def load_data(file_paths):
all_data = []
metric_keys = ['lin_vel_err', 'ang_vel_err', 'orientation_stability', 'dof_limits', 'torque_smoothness', 'dof_power']
labels_map = {
'lin_vel_err': 'Lin Vel\nAccuracy',
'ang_vel_err': 'Ang Vel\nAccuracy',
'dof_limits': 'Joint Limits\nMargin',
'dof_power': 'Energy\nEfficiency',
'orientation_stability': 'Orientation\nStability',
'torque_smoothness': 'Torque\nSmoothness',
}
for path in file_paths:
if not os.path.exists(path):
print(f"[ERROR] File not found: {path}")
continue
with open(path, 'r', encoding='utf-8') as f:
content = yaml.safe_load(f)
raw_path = content.get('model_path', 'Unknown_Model')
model_name = os.path.basename(raw_path).replace('.pt', '')
values = []
for k in metric_keys:
if k in content:
raw_val = content[k]['mean@50']
values.append(parse_value_string(raw_val))
else:
values.append(0.0)
all_data.append({'name': model_name, 'values': values})
return all_data, [labels_map[k] for k in metric_keys]
def plot_radar(data_list, labels, output_file=None, r_range=(0, 1.05)):
"""绘制雷达图"""
if not data_list:
return
num_vars = len(labels)
angles = np.linspace(0, 2 * np.pi, num_vars, endpoint=False).tolist()
angles += angles[:1]
colors = plt.cm.get_cmap("tab10", len(data_list))
fig, ax = plt.subplots(figsize=(10, 8), subplot_kw=dict(polar=True))
plt.subplots_adjust(left=0.05, right=0.75, top=0.9, bottom=0.1)
ax.set_theta_offset(np.pi / 2)
ax.set_theta_direction(-1)
plt.xticks(angles[:-1], labels, color='#444444', size=11)
for label, angle in zip(ax.get_xticklabels(), angles[:-1]):
if angle in (0, np.pi): label.set_horizontalalignment('center')
elif 0 < angle < np.pi: label.set_horizontalalignment('left')
else: label.set_horizontalalignment('right')
r_min, r_max = r_range
ax.set_ylim(r_min, r_max)
ticks = np.linspace(r_min, r_max, 5)
plt.yticks(ticks, [f"{t:.2f}" for t in ticks], color="grey", size=10)
ax.set_rlabel_position(180)
ax.grid(True, color='gray', linestyle=':', linewidth=1.5, alpha=0.5)
ax.spines['polar'].set_visible(False)
linewidth = 2.5
for idx, item in enumerate(data_list):
values_closed = item['values'] + [item['values'][0]]
color = colors(idx)
ax.plot(angles, values_closed, linewidth=linewidth, label=item['name'], color=color)
ax.fill(angles, values_closed, color=color, alpha=0.15)
plt.legend(loc='lower left', bbox_to_anchor=(0.82, 0.0), title="Models", frameon=False)
plt.title('Performance Comparison (Radar)', size=16, y=1.06)
if output_file:
plt.savefig(output_file, dpi=300, bbox_inches='tight')
print(f"Radar plot saved to {output_file}")
def plot_bar(data_list, labels, output_file=None, h_range=(0, 1.05)):
"""绘制柱状图"""
if not data_list:
return
num_models = len(data_list)
num_metrics = len(labels)
# 设置柱状图参数
x = np.arange(num_metrics)
width = 0.8 / num_models # 自动计算柱子宽度
fig, ax = plt.subplots(figsize=(12, 6))
colors = plt.cm.get_cmap("tab10", num_models)
for idx, item in enumerate(data_list):
# 计算每个模型柱子的偏移量
offset = (idx - (num_models - 1) / 2) * width
ax.bar(x + offset, item['values'], width, label=item['name'], color=colors(idx), alpha=0.8)
ax.set_ylabel('Score / Value')
ax.set_title('Performance Comparison (Bar)', size=16)
ax.set_xticks(x)
# 将标签中的换行符去掉或处理,使其在柱状图中更美观
clean_labels = [l.replace('\n', ' ') for l in labels]
ax.set_xticklabels(clean_labels, rotation=15, ha='right')
ax.legend(title="Models", bbox_to_anchor=(1.05, 1), loc='upper left', frameon=False)
ax.grid(axis='y', linestyle='--', alpha=0.7)
if h_range:
ax.set_ylim(h_range)
plt.tight_layout()
if output_file:
plt.savefig(output_file, dpi=300, bbox_inches='tight')
print(f"Bar plot saved to {output_file}")
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('files_or_dir', metavar='F', type=str, nargs='+', help='YAML files or directory')
parser.add_argument('--out', type=str, help='Base name for output files')
parser.add_argument('--range', type=float, nargs=2, default=[0.0, 1.0], help='Axis range for radar and bar plots')
args = parser.parse_args()
# 获取文件列表
files = args.files_or_dir
if Path(args.files_or_dir[0]).is_dir():
dir_path = Path(args.files_or_dir[0])
files = [str(p) for p in dir_path.glob('*.yaml')]
data, metrics_labels = load_data(files)
# 处理输出文件名
radar_out, bar_out = None, None
if args.out:
out_path = Path(args.out)
radar_out = str(out_path.with_name(f"{out_path.stem}_radar{out_path.suffix}"))
bar_out = str(out_path.with_name(f"{out_path.stem}_bar{out_path.suffix}"))
# 分别调用绘图函数
plot_radar(data, metrics_labels, output_file=radar_out, r_range=args.range)
plot_bar(data, metrics_labels, output_file=bar_out, h_range=args.range)
plt.show()