Codex Agent
712db9c597
Add B1 directory-type seed mechanism to AGENT_WORKSPACE_SEED_FILES: - Extended agentWorkspaceFilesForRun with collectDirectorySeedFiles helper - arm2d-skill (SKILL.md + references/ + python/) auto-injected to .agents/skills/arm2d-skill/ - Enables Code Agent to follow ARM-2D constraints and call asset scripts in case Skill source: https://github.com/notLabyet/HWLabOA
154 lines
6.3 KiB
Python
154 lines
6.3 KiB
Python
"""
|
||
视频抠图转序列帧工具 - 专为ARM2D动效/B站素材设计
|
||
========== 所有参数都在下面配置区改，不用动下面逻辑 ==========
|
||
"""
|
||
import warnings
|
||
warnings.filterwarnings("ignore") # 屏蔽无关性能警告
|
||
import cv2
|
||
from rembg import remove
|
||
from PIL import Image
|
||
import numpy as np
|
||
import os
|
||
|
||
# ====================== 配置区（你直接改这里就行） ======================
|
||
INPUT_VIDEO_PATH = r"C:\Users\11791\OneDrive\Desktop\images\girl.mp4" # 输入视频路径
|
||
OUTPUT_DIR = r"C:\Users\11791\OneDrive\Desktop\images\output_frames" # 输出帧保存路径
|
||
# 帧配置
|
||
TARGET_FRAME_COUNT = 100 # 要输出的总帧数，均匀从视频里抽取
|
||
TARGET_SIZE = (240, 240) # 输出图片尺寸（宽,高），和你LCD一致就行
|
||
AUTO_CROP_CENTER = True # 自动居中裁剪，避免拉伸变形，建议开
|
||
# 画质配置
|
||
CONTRAST = 1.0 # 对比度倍数，>1提高，视频暗就往大调
|
||
BRIGHTNESS = 10 # 亮度增加值，0~255，视频暗往大调
|
||
ENABLE_ALPHA_MATTING = False # 开边缘优化，抠出来无锯齿
|
||
ONLY_OUTPUT_MASK = False # 只输出黑白Alpha遮罩，做局部刷新时开
|
||
ENABLE_REMOVE_BG = True # 是否开启抠图，关闭后直接抽帧转GIF（速度提升10倍以上）
|
||
USE_GPU = False # 要是CUDA装好了就改成True，速度快10倍
|
||
# GIF输出配置
|
||
ENABLE_GIF_OUTPUT = True # 是否输出合并后的GIF
|
||
OUTPUT_GIF_PATH = r"C:\Users\11791\OneDrive\Desktop\images\output.gif" # GIF输出路径
|
||
GIF_FPS = 10 # GIF的播放帧率，10帧每秒就很流畅
|
||
GIF_LOOP_FOREVER = True # GIF是否无限循环
|
||
KEEP_FRAMES_AFTER_GIF = True # 生成GIF后是否保留单独的序列帧
|
||
# =======================================================================
|
||
|
||
# 自动创建输出目录
|
||
os.makedirs(OUTPUT_DIR, exist_ok=True)
|
||
# 配置校验
|
||
if not ENABLE_REMOVE_BG and ONLY_OUTPUT_MASK:
|
||
print("⚠️ 提示：不抠图模式下ONLY_OUTPUT_MASK选项无效，已自动关闭")
|
||
ONLY_OUTPUT_MASK = False
|
||
# 选择运行后端
|
||
providers = ['CUDAExecutionProvider', 'CPUExecutionProvider'] if USE_GPU else ['CPUExecutionProvider']
|
||
|
||
# 读取视频信息
|
||
cap = cv2.VideoCapture(INPUT_VIDEO_PATH, cv2.CAP_FFMPEG)
|
||
total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
|
||
# 计算要抽取的帧索引（均匀分布在整个视频）
|
||
extract_indices = [int(i * total_frames / TARGET_FRAME_COUNT) for i in range(TARGET_FRAME_COUNT)]
|
||
|
||
print(f"=== 开始处理 ===")
|
||
print(f"输入视频总帧数：{total_frames}")
|
||
print(f"将抽取{len(extract_indices)}帧，输出尺寸：{TARGET_SIZE[0]}×{TARGET_SIZE[1]}")
|
||
print(f"输出路径：{OUTPUT_DIR}\n")
|
||
|
||
for idx, frame_idx in enumerate(extract_indices):
|
||
# 跳到指定帧
|
||
cap.set(cv2.CAP_PROP_POS_FRAMES, frame_idx)
|
||
ret, frame = cap.read()
|
||
if not ret:
|
||
print(f"跳过第{idx+1}帧：读取失败")
|
||
continue
|
||
|
||
# 1. 调整对比度亮度
|
||
frame = cv2.convertScaleAbs(frame, alpha=CONTRAST, beta=BRIGHTNESS)
|
||
|
||
# 2. 裁剪+缩放
|
||
if AUTO_CROP_CENTER:
|
||
h, w = frame.shape[:2]
|
||
crop_size = min(w, h)
|
||
x_start = (w - crop_size) // 2
|
||
y_start = (h - crop_size) // 2
|
||
frame = frame[y_start:y_start+crop_size, x_start:x_start+crop_size]
|
||
frame = cv2.resize(frame, TARGET_SIZE)
|
||
|
||
# 3. 抠图/直接使用原图
|
||
frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
|
||
img_pil = Image.fromarray(frame_rgb)
|
||
if ENABLE_REMOVE_BG:
|
||
img_rembg = remove(
|
||
img_pil,
|
||
providers=providers,
|
||
alpha_matting=ENABLE_ALPHA_MATTING,
|
||
alpha_matting_discard_threshold=1e-5, # 优化矩阵计算，消除警告
|
||
alpha_matting_shift=1e-2,
|
||
only_mask=ONLY_OUTPUT_MASK
|
||
)
|
||
else:
|
||
# 不抠图模式，直接保留原图，添加全不透明alpha通道兼容后续逻辑
|
||
img_rembg = img_pil.convert("RGBA")
|
||
|
||
# 4. 保存
|
||
if ONLY_OUTPUT_MASK:
|
||
# 遮罩直接存灰度图
|
||
img_out = Image.fromarray(np.array(img_rembg) * 255).convert("L")
|
||
else:
|
||
# 存带透明通道的RGBA图
|
||
img_out = Image.fromarray(np.array(img_rembg))
|
||
|
||
output_path = os.path.join(OUTPUT_DIR, f"frame_{idx+1:02d}.png")
|
||
img_out.save(output_path)
|
||
print(f"已完成第{idx+1}/{TARGET_FRAME_COUNT}帧：{os.path.basename(output_path)}")
|
||
|
||
cap.release()
|
||
frame_count = len(os.listdir(OUTPUT_DIR))
|
||
print(f"\n=== 帧处理完成，共生成{frame_count}帧 ===")
|
||
|
||
# ========== 生成GIF ==========
|
||
if ENABLE_GIF_OUTPUT and frame_count > 0:
|
||
print("正在生成GIF...")
|
||
# 读取所有帧
|
||
frames = []
|
||
for i in range(1, frame_count+1):
|
||
frame_path = os.path.join(OUTPUT_DIR, f"frame_{i:02d}.png")
|
||
if os.path.exists(frame_path):
|
||
img = Image.open(frame_path)
|
||
# 处理透明通道，避免GIF透明异常
|
||
if img.mode == "RGBA":
|
||
# 转索引色，保留透明
|
||
alpha = img.getchannel("A")
|
||
img = img.convert("P", palette=Image.ADAPTIVE, colors=256)
|
||
mask = Image.eval(alpha, lambda a: 255 if a <= 128 else 0)
|
||
img.paste(255, mask=mask)
|
||
img.info["transparency"] = 255
|
||
frames.append(img)
|
||
|
||
if len(frames) > 0:
|
||
# 计算每帧延迟时间（毫秒）
|
||
frame_delay = int(1000 / GIF_FPS)
|
||
loop = 0 if GIF_LOOP_FOREVER else 1
|
||
# 保存GIF
|
||
frames[0].save(
|
||
OUTPUT_GIF_PATH,
|
||
save_all=True,
|
||
append_images=frames[1:],
|
||
duration=frame_delay,
|
||
loop=loop,
|
||
disposal=2, # 每帧播放完清除，避免透明残留
|
||
optimize=False
|
||
)
|
||
print(f"GIF生成成功：{OUTPUT_GIF_PATH}")
|
||
|
||
# 删除单独帧（如果配置了）
|
||
if not KEEP_FRAMES_AFTER_GIF:
|
||
for f in os.listdir(OUTPUT_DIR):
|
||
os.remove(os.path.join(OUTPUT_DIR, f))
|
||
os.rmdir(OUTPUT_DIR)
|
||
print("已清理临时序列帧文件")
|
||
|
||
print(f"\n=== 全部处理完成 ===")
|
||
if KEEP_FRAMES_AFTER_GIF or not ENABLE_GIF_OUTPUT:
|
||
print(f"序列帧保存在：{OUTPUT_DIR}")
|
||
if ENABLE_GIF_OUTPUT:
|
||
print(f"GIF保存在：{OUTPUT_GIF_PATH}")
|