from pathlib import Path
import argparse
import json
import math, cv2, numpy as np, onnxruntime as ort
from numpy import ndarray
from pydantic import BaseModel


class BoundingBox(BaseModel):
    x1: int
    y1: int
    x2: int
    y2: int
    cls_id: int
    conf: float


class TVFrameResult(BaseModel):
    frame_id: int
    boxes: list[BoundingBox]
    keypoints: list[tuple[int, int]]


class Miner:
    def __init__(self, path_hf_repo: Path) -> None:
        model_path = path_hf_repo / "weights.onnx"

        try:
            ort.preload_dlls()
        except Exception:
            pass

        sess_options = ort.SessionOptions()
        sess_options.graph_optimization_level = ort.GraphOptimizationLevel.ORT_ENABLE_ALL

        try:
            self.session = ort.InferenceSession(
                str(model_path),
                sess_options=sess_options,
                providers=["CUDAExecutionProvider", "CPUExecutionProvider"],
            )
        except Exception:
            self.session = ort.InferenceSession(
                str(model_path),
                sess_options=sess_options,
                providers=["CPUExecutionProvider"],
            )

        self.input_name = self.session.get_inputs()[0].name
        self.output_names = [o.name for o in self.session.get_outputs()]
        self.input_shape = self.session.get_inputs()[0].shape

        self.input_height = self._safe_dim(self.input_shape[2], 1280)
        self.input_width = self._safe_dim(self.input_shape[3], 1280)

        self.tile_size = 960
        self.overlap = 0.5

        self.conf_thres = 0.47
        self.wbf_iou = 0.55
        self.max_det = 100

        self.min_box_area = 14 * 14
        self.min_w = 8
        self.min_h = 8
        self.max_aspect_ratio = 8.0
        self.max_box_area_ratio = 0.8

    def __repr__(self) -> str:
        return (
            f"ONNXRuntime(session={type(self.session).__name__}, "
            f"providers={self.session.get_providers()})"
        )

    @staticmethod
    def _safe_dim(value, default: int) -> int:
        return value if isinstance(value, int) and value > 0 else default

    @staticmethod
    def _clip_boxes(boxes: np.ndarray, image_size: tuple[int, int]) -> np.ndarray:
        w, h = image_size
        if len(boxes) == 0:
            return boxes
        boxes[:, 0] = np.clip(boxes[:, 0], 0, w - 1)
        boxes[:, 1] = np.clip(boxes[:, 1], 0, h - 1)
        boxes[:, 2] = np.clip(boxes[:, 2], 0, w - 1)
        boxes[:, 3] = np.clip(boxes[:, 3], 0, h - 1)
        return boxes

    @staticmethod
    def _xywh_to_xyxy(boxes: np.ndarray) -> np.ndarray:
        out = np.empty_like(boxes)
        out[:, 0] = boxes[:, 0] - boxes[:, 2] / 2.0
        out[:, 1] = boxes[:, 1] - boxes[:, 3] / 2.0
        out[:, 2] = boxes[:, 0] + boxes[:, 2] / 2.0
        out[:, 3] = boxes[:, 1] + boxes[:, 3] / 2.0
        return out

    def _slice_image(
        self, image: np.ndarray
    ) -> list[tuple[np.ndarray, tuple[int, int], tuple[int, int]]]:
        h, w = image.shape[:2]
        t = self.tile_size
        st = max(1, int(t * (1.0 - self.overlap)))

        xs = []
        x = 0
        while True:
            if x + t >= w:
                xs.append(max(0, w - t))
                break
            xs.append(x)
            x += st

        ys = []
        y = 0
        while True:
            if y + t >= h:
                ys.append(max(0, h - t))
                break
            ys.append(y)
            y += st

        xs = list(dict.fromkeys(xs))
        ys = list(dict.fromkeys(ys))

        out = []
        for y0 in ys:
            for x0 in xs:
                x1 = min(x0 + t, w)
                y1 = min(y0 + t, h)
                crop = image[y0:y1, x0:x1]
                vh, vw = crop.shape[:2]
                out.append((crop, (x0, y0), (vw, vh)))
        return out

    def _preprocess(self, image: np.ndarray) -> np.ndarray:
        if image.dtype != np.uint8:
            image = image.astype(np.uint8)

        h, w = image.shape[:2]
        pad_w = max(0, self.input_width - w)
        pad_h = max(0, self.input_height - h)

        if pad_w > 0 or pad_h > 0:
            image = cv2.copyMakeBorder(
                image,
                0,
                pad_h,
                0,
                pad_w,
                cv2.BORDER_CONSTANT,
                value=(114, 114, 114),
            )

        if image.shape[0] != self.input_height or image.shape[1] != self.input_width:
            raise ValueError(
                f"Bad tile shape after pad: {image.shape[:2]}, "
                f"expected ({self.input_height}, {self.input_width})"
            )

        img = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
        img = img.astype(np.float32) / 255.0
        img = np.transpose(img, (2, 0, 1))[None, ...]
        return np.ascontiguousarray(img, dtype=np.float32)

    def _filter_sane_boxes(
        self,
        boxes: np.ndarray,
        scores: np.ndarray,
        cls_ids: np.ndarray,
        valid_size: tuple[int, int],
    ) -> tuple[np.ndarray, np.ndarray, np.ndarray]:
        if len(boxes) == 0:
            return boxes, scores, cls_ids

        valid_w, valid_h = valid_size
        image_area = float(max(1, valid_w * valid_h))

        keep = []
        for i, box in enumerate(boxes):
            x1, y1, x2, y2 = box.tolist()
            bw = x2 - x1
            bh = y2 - y1

            if bw <= 0 or bh <= 0:
                continue
            if bw < self.min_w or bh < self.min_h:
                continue

            area = bw * bh
            if area < self.min_box_area:
                continue
            if area > self.max_box_area_ratio * image_area:
                continue

            ar = max(bw / max(bh, 1e-6), bh / max(bw, 1e-6))
            if ar > self.max_aspect_ratio:
                continue

            keep.append(i)

        if not keep:
            return (
                np.empty((0, 4), dtype=np.float32),
                np.empty((0,), dtype=np.float32),
                np.empty((0,), dtype=np.int32),
            )

        keep = np.array(keep, dtype=np.intp)
        return boxes[keep], scores[keep], cls_ids[keep]

    def _decode_final_dets(
        self, preds: np.ndarray, valid_size: tuple[int, int]
    ) -> tuple[np.ndarray, np.ndarray]:
        if preds.ndim == 3 and preds.shape[0] == 1:
            preds = preds[0]

        if preds.ndim != 2 or preds.shape[1] < 6:
            raise ValueError(f"Unexpected final-det output shape: {preds.shape}")

        boxes = preds[:, :4].astype(np.float32)
        scores = preds[:, 4].astype(np.float32)
        cls_ids = preds[:, 5].astype(np.int32)

        keep = cls_ids == 0
        boxes = boxes[keep]
        scores = scores[keep]
        cls_ids = cls_ids[keep]

        keep = scores >= self.conf_thres
        boxes = boxes[keep]
        scores = scores[keep]
        cls_ids = cls_ids[keep]

        if len(boxes) == 0:
            return (
                np.empty((0, 4), dtype=np.float32),
                np.empty((0,), dtype=np.float32),
            )

        boxes = self._clip_boxes(boxes, valid_size)
        boxes, scores, cls_ids = self._filter_sane_boxes(boxes, scores, cls_ids, valid_size)

        if len(boxes) == 0:
            return (
                np.empty((0, 4), dtype=np.float32),
                np.empty((0,), dtype=np.float32),
            )

        return boxes, scores

    def _decode_raw_yolo(
        self, preds: np.ndarray, valid_size: tuple[int, int]
    ) -> tuple[np.ndarray, np.ndarray]:
        if preds.ndim != 3:
            raise ValueError(f"Unexpected raw output shape: {preds.shape}")
        if preds.shape[0] != 1:
            raise ValueError(f"Unexpected batch dimension in raw output: {preds.shape}")

        preds = preds[0]

        if preds.shape[0] <= 16 and preds.shape[1] > preds.shape[0]:
            preds = preds.T

        if preds.ndim != 2 or preds.shape[1] < 5:
            raise ValueError(f"Unexpected normalized raw output shape: {preds.shape}")

        boxes_xywh = preds[:, :4].astype(np.float32)
        tail = preds[:, 4:].astype(np.float32)

        if tail.shape[1] == 1:
            scores = tail[:, 0]
            cls_ids = np.zeros(len(scores), dtype=np.int32)
        elif tail.shape[1] == 2:
            obj = tail[:, 0]
            cls_prob = tail[:, 1]
            scores = obj * cls_prob
            cls_ids = np.zeros(len(scores), dtype=np.int32)
        else:
            obj = tail[:, 0]
            class_probs = tail[:, 1:]
            cls_ids = np.argmax(class_probs, axis=1).astype(np.int32)
            cls_scores = class_probs[np.arange(len(class_probs)), cls_ids]
            scores = obj * cls_scores

        keep = cls_ids == 0
        boxes_xywh = boxes_xywh[keep]
        scores = scores[keep]
        cls_ids = cls_ids[keep]

        keep = scores >= self.conf_thres
        boxes_xywh = boxes_xywh[keep]
        scores = scores[keep]
        cls_ids = cls_ids[keep]

        if len(boxes_xywh) == 0:
            return (
                np.empty((0, 4), dtype=np.float32),
                np.empty((0,), dtype=np.float32),
            )

        boxes = self._xywh_to_xyxy(boxes_xywh)
        boxes = self._clip_boxes(boxes, valid_size)
        boxes, scores, cls_ids = self._filter_sane_boxes(boxes, scores, cls_ids, valid_size)

        if len(boxes) == 0:
            return (
                np.empty((0, 4), dtype=np.float32),
                np.empty((0,), dtype=np.float32),
            )

        return boxes, scores

    def _postprocess(
        self, output: np.ndarray, valid_size: tuple[int, int]
    ) -> tuple[np.ndarray, np.ndarray]:
        if output.ndim == 2 and output.shape[1] == 6:
            return self._decode_final_dets(output, valid_size)

        if output.ndim == 3 and output.shape[0] == 1 and output.shape[2] == 6:
            return self._decode_final_dets(output, valid_size)

        return self._decode_raw_yolo(output, valid_size)

    def _predict_tile(self, image: np.ndarray) -> tuple[np.ndarray, np.ndarray]:
        if image is None:
            raise ValueError("Input image is None")
        if not isinstance(image, np.ndarray):
            raise TypeError(f"Input is not numpy array: {type(image)}")
        if image.ndim != 3:
            raise ValueError(f"Expected HWC image, got shape={image.shape}")
        if image.shape[2] != 3:
            raise ValueError(f"Expected 3 channels, got shape={image.shape}")

        valid_w, valid_h = image.shape[1], image.shape[0]
        input_tensor = self._preprocess(image)

        expected_shape = (1, 3, self.input_height, self.input_width)
        if input_tensor.shape != expected_shape:
            raise ValueError(
                f"Bad input tensor shape={input_tensor.shape}, expected={expected_shape}"
            )

        outputs = self.session.run(self.output_names, {self.input_name: input_tensor})
        det_output = outputs[0]
        return self._postprocess(det_output, (valid_w, valid_h))

    @staticmethod
    def _iou_one(a: np.ndarray, b: np.ndarray) -> float:
        xx1 = max(float(a[0]), float(b[0]))
        yy1 = max(float(a[1]), float(b[1]))
        xx2 = min(float(a[2]), float(b[2]))
        yy2 = min(float(a[3]), float(b[3]))

        inter = max(0.0, xx2 - xx1) * max(0.0, yy2 - yy1)
        area_a = max(0.0, float(a[2] - a[0])) * max(0.0, float(a[3] - a[1]))
        area_b = max(0.0, float(b[2] - b[0])) * max(0.0, float(b[3] - b[1]))
        return float(inter / (area_a + area_b - inter + 1e-7))

    def _wbf(self, boxes: np.ndarray, scores: np.ndarray) -> tuple[np.ndarray, np.ndarray]:
        if len(boxes) == 0:
            return boxes, scores

        order = np.argsort(scores)[::-1]
        boxes = boxes[order]
        scores = scores[order]

        used = np.zeros(len(boxes), dtype=bool)
        fused_boxes = []
        fused_scores = []

        for i in range(len(boxes)):
            if used[i]:
                continue

            cluster = [i]
            used[i] = True
            changed = True

            while changed:
                changed = False
                for j in range(i + 1, len(boxes)):
                    if used[j]:
                        continue
                    if any(self._iou_one(boxes[j], boxes[k]) >= self.wbf_iou for k in cluster):
                        used[j] = True
                        cluster.append(j)
                        changed = True

            cb = boxes[cluster]
            cs = scores[cluster]
            w = cs / (cs.sum() + 1e-7)

            fused_boxes.append((cb * w[:, None]).sum(axis=0))
            fused_scores.append(float(cs.max()))

        return np.asarray(fused_boxes, dtype=np.float32), np.asarray(fused_scores, dtype=np.float32)

    def _merge_boxes_global(
        self,
        image_shape: tuple[int, int],
        tile_boxes: np.ndarray,
        tile_scores: np.ndarray,
    ) -> list[BoundingBox]:
        if len(tile_boxes) == 0:
            return []

        H, W = image_shape[:2]
        boxes = tile_boxes.copy()
        scores = tile_scores.copy()

        keep = np.argsort(scores)[::-1]
        boxes = boxes[keep]
        scores = scores[keep]

        boxes, scores = self._wbf(boxes, scores)

        out = []
        for b, sc in zip(boxes[: self.max_det], scores[: self.max_det]):
            x1 = int(math.floor(b[0]))
            y1 = int(math.floor(b[1]))
            x2 = int(math.ceil(b[2]))
            y2 = int(math.ceil(b[3]))

            x1 = max(0, min(x1, W - 1))
            y1 = max(0, min(y1, H - 1))
            x2 = max(0, min(x2, W - 1))
            y2 = max(0, min(y2, H - 1))

            if x2 > x1 and y2 > y1:
                out.append(
                    BoundingBox(
                        x1=x1,
                        y1=y1,
                        x2=x2,
                        y2=y2,
                        cls_id=0,
                        conf=float(sc),
                    )
                )
        return out

    def predict_image(self, image: np.ndarray) -> list[BoundingBox]:
        if image is None:
            raise ValueError("Input image is None")
        if not isinstance(image, np.ndarray):
            raise TypeError(f"Input is not numpy array: {type(image)}")
        if image.ndim != 3:
            raise ValueError(f"Expected HWC image, got shape={image.shape}")
        if image.shape[2] != 3:
            raise ValueError(f"Expected 3 channels, got shape={image.shape}")

        H, W = image.shape[:2]
        tiles = self._slice_image(image)

        all_boxes = []
        all_scores = []

        for i, (tile_img, (ox, oy), (vw, vh)) in enumerate(tiles):
            try:
                b1, s1 = self._predict_tile(tile_img)

                flipped = cv2.flip(tile_img, 1)
                b2, s2 = self._predict_tile(flipped)

                if len(b2):
                    tw = tile_img.shape[1]
                    b2 = b2.copy()
                    b2[:, [0, 2]] = tw - b2[:, [2, 0]]

                if len(b1) == 0 and len(b2) == 0:
                    continue
                if len(b1) == 0:
                    b, s = b2, s2
                elif len(b2) == 0:
                    b, s = b1, s1
                else:
                    b = np.vstack([b1, b2])
                    s = np.hstack([s1, s2])

            except Exception as e:
                print(f"⚠️ Tile inference failed at ({ox}, {oy}): {e}")
                continue

            left_edge = ox == 0
            top_edge = oy == 0
            right_edge = (ox + vw) >= W
            bottom_edge = (oy + vh) >= H

            for bb, sc in zip(b, s):
                bw = bb[2] - bb[0]
                bh = bb[3] - bb[1]
                m = max(8, int(min(bw, bh) * 0.2))

                if not left_edge and bb[0] < m:
                    continue
                if not top_edge and bb[1] < m:
                    continue
                if not right_edge and bb[2] > (vw - m):
                    continue
                if not bottom_edge and bb[3] > (vh - m):
                    continue

                x1 = max(0, min(W - 1, int(bb[0] + ox)))
                y1 = max(0, min(H - 1, int(bb[1] + oy)))
                x2 = max(0, min(W - 1, int(bb[2] + ox)))
                y2 = max(0, min(H - 1, int(bb[3] + oy)))

                if x2 > x1 and y2 > y1:
                    all_boxes.append([x1, y1, x2, y2])
                    all_scores.append(float(sc))

        if not all_boxes:
            return []

        boxes = np.asarray(all_boxes, dtype=np.float32)
        scores = np.asarray(all_scores, dtype=np.float32)

        return self._merge_boxes_global(image.shape, boxes, scores)

    def predict_batch(
        self,
        batch_images: list[ndarray],
        offset: int,
        n_keypoints: int,
    ) -> list[TVFrameResult]:
        results: list[TVFrameResult] = []

        for frame_number_in_batch, image in enumerate(batch_images):
            try:
                boxes = self.predict_image(image)
            except Exception as e:
                print(f"⚠️ Inference failed for frame {offset + frame_number_in_batch}: {e}")
                boxes = []

            results.append(
                TVFrameResult(
                    frame_id=offset + frame_number_in_batch,
                    boxes=boxes,
                    keypoints=[(0, 0) for _ in range(max(0, int(n_keypoints)))],
                )
            )

        return results