BatchRaycast_2d.py

import torch

def bresenham_line(x0: int, y0: int, x1: int, y1: int):
    """
    Integer Bresenham line algorithm.
    Returns two Python lists: xs, ys (same length).
    """
    xs = []
    ys = []

    dx = abs(x1 - x0)
    sx = 1 if x0 < x1 else -1
    dy = -abs(y1 - y0)
    sy = 1 if y0 < y1 else -1
    err = dx + dy  # error value e_xy

    x, y = x0, y0
    while True:
        xs.append(x)
        ys.append(y)
        if x == x1 and y == y1:
            break
        e2 = 2 * err
        if e2 >= dy:
            err += dy
            x += sx
        if e2 <= dx:
            err += dx
            y += sy

    return xs, ys


class BatchRaycast_2D:
    """
    Returns the first image in the batch where the START->END line
    is completely "white enough" according to the chosen mode/threshold.
    """

    @classmethod
    def INPUT_TYPES(cls):
        return {
            "required": {
                "images": ("IMAGE",),

                # "White" detection behavior:
                # - max_channel: max(R,G,B) >= threshold  (VERY tolerant; rgb(0,1,0) passes)
                # - all_channels: min(R,G,B) >= threshold (strict white)
                # - luminance: dot([0.2126, 0.7152, 0.0722], RGB) >= threshold
                # - green_only: G >= threshold
                "white_mode": (["max_channel", "all_channels", "luminance", "green_only"], {"default": "max_channel"}),

                # For ComfyUI IMAGE tensors this is typically 0..1 float.
                # Example: threshold=0.98 means "near 1.0"
                "threshold": ("FLOAT", {"default": 0.98, "min": 0.0, "max": 1.0, "step": 0.005}),

                # What to do if no image matches:
                "fallback": (["return_first", "return_last"], {"default": "return_last"}),
            },
            "optional": {
                # Keep your requested defaults, but allow override if needed.
                "start_x": ("INT", {"default": 0, "min": -999999, "max": 999999}),
                "start_y": ("INT", {"default": 386, "min": -999999, "max": 999999}),
                "end_x": ("INT", {"default": 330, "min": -999999, "max": 999999}),
                "end_y": ("INT", {"default": 385, "min": -999999, "max": 999999}),
            }
        }

    RETURN_TYPES = ("IMAGE", "INT")
    RETURN_NAMES = ("image", "index")
    FUNCTION = "pick"
    CATEGORY = "image/filter"

    def pick(
        self,
        images,
        white_mode="max_channel",
        threshold=0.98,
        fallback="return_last",
        start_x=0,
        start_y=386,
        end_x=330,
        end_y=385,
    ):
        # images: torch tensor [B,H,W,C], float in 0..1 (typical in ComfyUI)
        if not isinstance(images, torch.Tensor):
            raise TypeError("images must be a torch.Tensor (ComfyUI IMAGE type).")

        if images.ndim != 4 or images.shape[-1] < 3:
            raise ValueError(f"Expected images shape [B,H,W,C>=3], got {tuple(images.shape)}")

        B, H, W, C = images.shape

        # Build the integer pixel coordinates along the line
        xs_list, ys_list = bresenham_line(int(start_x), int(start_y), int(end_x), int(end_y))

        # Convert to tensors on same device
        device = images.device
        xs = torch.tensor(xs_list, device=device, dtype=torch.long)
        ys = torch.tensor(ys_list, device=device, dtype=torch.long)

        # If the line goes out of bounds, we treat it as "no match" for safety.
        if xs.min().item() < 0 or ys.min().item() < 0 or xs.max().item() >= W or ys.max().item() >= H:
            # fallback output
            idx = 0 if fallback == "return_first" else max(B - 1, 0)
            return (images[idx:idx+1], int(-1))

        # Sample pixels along the line for every image in the batch: shape [B, N, 3]
        pixels = images[:, ys, xs, :3]

        if white_mode == "max_channel":
            # VERY tolerant: any channel close to 1 counts as "white"
            white_mask = pixels.max(dim=-1).values >= threshold

        elif white_mode == "all_channels":
            # Strict white: all channels must be close to 1
            white_mask = pixels.min(dim=-1).values >= threshold

        elif white_mode == "green_only":
            # Only green channel must be high
            white_mask = pixels[..., 1] >= threshold

        elif white_mode == "luminance":
            # Perceived brightness (Rec.709-ish)
            weights = torch.tensor([0.2126, 0.7152, 0.0722], device=device, dtype=pixels.dtype)
            lum = (pixels * weights).sum(dim=-1)
            white_mask = lum >= threshold

        else:
            raise ValueError(f"Unknown white_mode: {white_mode}")

        # Line is "completely white" if ALL pixels along line are white
        line_is_white = white_mask.all(dim=1)  # shape [B]

        found = torch.nonzero(line_is_white, as_tuple=False).flatten()
        if found.numel() > 0:
            idx = int(found[0].item())
            return (images[idx:idx+1], idx)

        # No match: fallback
        idx = 0 if fallback == "return_first" else max(B - 1, 0)
        return (images[idx:idx+1], int(-1))


NODE_CLASS_MAPPINGS = {
    "BatchRaycast_2D": BatchRaycast_2D
}

NODE_DISPLAY_NAME_MAPPINGS = {
    "BatchRaycast_2D": "BatchRaycast_2D"
}