hibana2077's picture
add data processing scripts and ffhq dataset; implement image-label mapping and visualization
548d832
import os
import sys
import pickle
from PIL import Image
import random
import matplotlib.pyplot as plt
# load the data(ffhq_data.pkl)
with open("ffhq_data.pkl", "rb") as f:
data = pickle.load(f)
# 假設 data 已經從 ffhq_data.pkl 載入
# 隨機抽取 5 張圖片
selected_data = random.sample(data, 5)
# 建立 1x5 的子圖區域,調整畫布大小以避免文字重疊
fig, axes = plt.subplots(1, 5, figsize=(20, 4))
for ax, item in zip(axes, selected_data):
img = item['img'] # PIL 圖片物件
label = item['label'] # 標籤字典
# 分離 headPose 與其他標籤資料
head_pose = label.get("headPose", None)
other_labels = {k: v for k, v in label.items() if k != "headPose"}
# 顯示圖片
ax.imshow(img)
ax.axis('off')
# 格式化 headPose 文字(若存在),並以多行呈現
if head_pose:
head_pose_text = "headPose:\n" + "\n".join([f"{k}: {v}" for k, v in head_pose.items()])
else:
head_pose_text = ""
# 格式化其他標籤文字
other_text = "\n".join([f"{k}: {v}" for k, v in other_labels.items()])
# 將 headPose 文字顯示在圖片上方(利用 ax.set_title)
if head_pose_text:
ax.set_title(head_pose_text, fontsize=10, pad=10)
# 將其他標籤文字顯示在圖片下方
ax.text(0.5, -0.1, other_text, transform=ax.transAxes, fontsize=10,
ha='center', va='top')
plt.tight_layout()
plt.savefig("output.png", bbox_inches='tight')