使用Numba加速纯Python嵌套列表操作时遇编译错误求助

问题背景 #

我被指派实现基于图像金字塔的纯Python图像融合程序，不可使用NumPy但可使用Numba。编写的图像模糊函数中，嵌套列表操作按Numba官方文档应为支持项，但运行时出现编译错误。以下是代码、高斯核定义及报错信息：

模糊函数代码 #

@jit
def blur_image(image: list[list[list[float], kernel: list[list[float]]) -> list[list[list[float]:
    PADDING: int = 2
    height: int = len(image)
    width: int = len(image[0])
    padded_img= create_empty_image(height + 2 * PADDING, width + 2 * PADDING)
    output_img = create_empty_image(height, width)

    for i in range(height):
        for j in range(width):
            padded_img[i + PADDING][j + PADDING] = image[i][j]

    for i in range(height):
        for j in range(width):
            sum_r: float = 0.
            sum_g: float = 0.
            sum_b: float = 0.
            for k in range(5):
                for l in range(5):
                    sum_r += kernel[k][l] * padded_img[i + k][j + l][0]
                    sum_g += kernel[k][l] * padded_img[i + k][j + l][1]
                    sum_b += kernel[k][l] * padded_img[i + k][j + l][2]

            output_img[i][j][0] = min(int(sum_r), 255)
            output_img[i][j][1] = min(int(sum_g), 255)
            output_img[i][j][2] = min(int(sum_b), 255)

    return output_img

@jit
def create_empty_image(height: int, width: int) -> img_as_list:
    return [[[0., 0., 0.] for _ in range(width)] for _ in range(height)]

高斯核定义 #

gaussian_kernel: list[list[float]] = [[1, 4, 6, 4, 1],
                                      [4, 16, 24, 16, 4],
                                      [6, 24, 36, 24, 6],
                                      [4, 16, 24, 16, 4],
                                      [1, 4, 6, 4, 1]] # * (1/256.), 归一化操作在其他地方完成

报错信息 #

Traceback (most recent call last):
  File "/home/krzysztof/lab5/lab_5.py", line 258, in <module>
    orange_upscaled = blur_image(orange_copy, gaussian_kernel)
  File "/home/krzysztof/venv/lib/python3.10/site-packages/numba/core/dispatcher.py", line 487, in _compile_for_args
    raise e
  File "/home/krzysztof/venv/lib/python3.10/site-packages/numba/core/dispatcher.py", line 420, in _compile_for_args
    return_val = self.compile(tuple(argtypes))
  File "/home/krzysztof/venv/lib/python3.10/site-packages/numba/core/dispatcher.py", line 1197, in compile
    cres = compiler.compile_ir(typingctx=self.typingctx,
  File "/home/krzysztof/venv/lib/python3.10/site-packages/numba/core/compiler.py", line 754, in compile_ir
    norw_cres = compile_local(func_ir.copy(), norw_flags)
  File "/home/krzysztof/venv/lib/python3.10/site-packages/numba/core/compiler.py", line 750, in compile_local
    return pipeline.compile_ir(func_ir=the_ir, lifted=lifted,
  File "/home/krzysztof/venv/lib/python3.10/site-packages/numba/core/compiler.py", line 462, in compile_ir
    return self._compile_ir()
  File "/home/krzysztof/venv/lib/python3.10/site-packages/numba/core/compiler.py", line 527, in _compile_ir
    return self._compile_core()
  File "/home/krzysztof/venv/lib/python3.10/site-packages/numba/core/compiler.py", line 499, in _compile_core
    raise e
  File "/home/krzysztof/venv/lib/python3.10/site-packages/numba/core/compiler.py", line 486, in _compile_core
    pm.run(self.state)
  File "/home/krzysztof/venv/lib/python3.10/site-packages/numba/core/compiler_machinery.py", line 368, in run
    raise patched_exception
  File "/home/krzysztof/venv/lib/python3.10/site-packages/numba/core/compiler_machinery.py", line 356, in run
    self._runPass(idx, pass_inst, state)
  File "/home/krzysztof/venv/lib/python3.10/site-packages/numba/core/compiler_lock.py", line 35, in _acquire_compile_lock
    return func(*args, **kwargs)
  File "/home/krzysztof/venv/lib/python3.10/site-packages/numba/core/compiler_machinery.py", line 318, in _runPass
    enforce_no_dels(internal_state.func_ir)
  File "/home/krzysztof/venv/lib/python3.10/site-packages/numba/core/ir_utils.py", line 2194, in enforce_no_dels
    raise CompilerError(msg, loc=dels[0].loc)
numba.core.errors.CompilerError: Failed in object mode pipeline (step: remove phis nodes)
Illegal IR, del found at: del $158for_iter.3

File "lab_5.py", line 204:
def blur_image(image: img_as_list, kernel: list[list[float]]) -> img_as_list:
    <source elided>
            sum_b: float = 0.
            for k in range(5):
            ^

Process finished with exit code 1

解决方法 #

1. 修复函数签名语法错误 #

原代码中blur_image的类型注解存在括号不匹配问题，导致语法解析异常，间接引发Numba编译错误。正确签名如下：

@njit
def blur_image(image: list[list[list[float]]], kernel: list[list[float]]) -> list[list[list[float]]]:

2. 启用Numba的nopython模式 #

用@njit（等价于@jit(nopython=True)）替代@jit，强制进入nopython编译模式。该模式性能更高，且能规避object模式下的IR生成问题。

3. 修正返回类型注解 #

create_empty_image的返回类型img_as_list若为自定义类型，需确保Numba能识别；若未定义，直接使用list[list[list[float]]]即可。

修正后的完整代码 #

from numba import njit

@njit
def blur_image(image: list[list[list[float]]], kernel: list[list[float]]) -> list[list[list[float]]]:
    PADDING: int = 2
    height: int = len(image)
    width: int = len(image[0])
    padded_img = create_empty_image(height + 2 * PADDING, width + 2 * PADDING)
    output_img = create_empty_image(height, width)

    # 填充图像边界
    for i in range(height):
        for j in range(width):
            padded_img[i + PADDING][j + PADDING] = image[i][j]

    # 高斯卷积计算
    for i in range(height):
        for j in range(width):
            sum_r = 0.0
            sum_g = 0.0
            sum_b = 0.0
            for k in range(5):
                for l in range(5):
                    sum_r += kernel[k][l] * padded_img[i + k][j + l][0]
                    sum_g += kernel[k][l] * padded_img[i + k][j + l][1]
                    sum_b += kernel[k][l] * padded_img[i + k][j + l][2]

            # 限制像素值在0-255范围内
            output_img[i][j][0] = min(int(sum_r), 255)
            output_img[i][j][1] = min(int(sum_g), 255)
            output_img[i][j][2] = min(int(sum_b), 255)

    return output_img

@njit
def create_empty_image(height: int, width: int) -> list[list[list[float]]]:
    return [[[0.0, 0.0, 0.0] for _ in range(width)] for _ in range(height)]

额外提示 #

确保高斯核已完成归一化（乘以1/256），否则卷积后的像素值可能超出合理范围，导致图像失真。

内容的提问来源于stack exchange，提问作者KrzysiekDD