ailai_image_point_diff/detect_image/bag_detect.py

import os
import cv2
import numpy as np
from rknnlite.api import RKNNLite

# ====================== 配置 ======================
MODEL_PATH = "bag3588.rknn"      # RKNN 模型路径
IMG_PATH = "2.jpg"     # 待推理图片路径
IMG_SIZE = (640, 640)           # 模型输入尺寸 (w,h)
OBJ_THRESH = 0.001              # 目标置信度阈值
NMS_THRESH = 0.45               # NMS 阈值
CLASS_NAME = ["bag"]
OUTPUT_DIR = "./result"
os.makedirs(OUTPUT_DIR, exist_ok=True)

# ====================== 工具函数 ======================
def letterbox_resize(image, size, bg_color=114):
    target_w, target_h = size
    h, w = image.shape[:2]
    scale = min(target_w / w, target_h / h)
    new_w, new_h = int(w * scale), int(h * scale)
    resized = cv2.resize(image, (new_w, new_h))
    canvas = np.full((target_h, target_w, 3), bg_color, dtype=np.uint8)
    dx, dy = (target_w - new_w) // 2, (target_h - new_h) // 2
    canvas[dy:dy + new_h, dx:dx + new_w] = resized
    return canvas, scale, dx, dy

def dfl_numpy(position):
    n, c, h, w = position.shape
    p_num = 4
    mc = c // p_num
    y = position.reshape(n, p_num, mc, h, w)
    y = np.exp(y) / np.sum(np.exp(y), axis=2, keepdims=True)
    acc = np.arange(mc).reshape(1,1,mc,1,1)
    y = np.sum(y * acc, axis=2)
    return y

def box_process(position):
    grid_h, grid_w = position.shape[2:4]
    col, row = np.meshgrid(np.arange(grid_w), np.arange(grid_h))
    col = col.reshape(1,1,grid_h,grid_w)
    row = row.reshape(1,1,grid_h,grid_w)
    grid = np.concatenate((col,row), axis=1)
    stride = np.array([IMG_SIZE[1] // grid_h, IMG_SIZE[0] // grid_w]).reshape(1,2,1,1)
    position = dfl_numpy(position)
    box_xy = grid + 0.5 - position[:,0:2,:,:]
    box_xy2 = grid + 0.5 + position[:,2:4,:,:]
    xyxy = np.concatenate((box_xy*stride, box_xy2*stride), axis=1)
    return xyxy

def filter_boxes(boxes, box_confidences, box_class_probs):
    boxes = np.array(boxes).reshape(-1, 4)
    box_confidences = np.array(box_confidences).reshape(-1)
    box_class_probs = np.array(box_class_probs)
    
    class_ids = np.argmax(box_class_probs, axis=-1)
    class_scores = box_class_probs[np.arange(len(class_ids)), class_ids]
    scores = box_confidences * class_scores

    mask = scores >= OBJ_THRESH
    if np.sum(mask) == 0:
        return None, None, None, None

    boxes = boxes[mask]
    classes = class_ids[mask]
    scores = scores[mask]
    conf_keep = box_confidences[mask]   # 原始 objectness

    # NMS
    x1, y1, x2, y2 = boxes[:,0], boxes[:,1], boxes[:,2], boxes[:,3]
    areas = (x2 - x1 + 1) * (y2 - y1 + 1)
    order = scores.argsort()[::-1]
    keep = []
    while order.size > 0:
        i = order[0]
        keep.append(i)
        xx1 = np.maximum(x1[i], x1[order[1:]])
        yy1 = np.maximum(y1[i], y1[order[1:]])
        xx2 = np.minimum(x2[i], x2[order[1:]])
        yy2 = np.minimum(y2[i], y2[order[1:]])
        w = np.maximum(0, xx2 - xx1 + 1)
        h = np.maximum(0, yy2 - yy1 + 1)
        inter = w * h
        ovr = inter / (areas[i] + areas[order[1:]] - inter)
        inds = np.where(ovr <= NMS_THRESH)[0]
        order = order[inds + 1]
    return boxes[keep], classes[keep], scores[keep], conf_keep[keep]

def post_process(outputs, scale, dx, dy):
    boxes_list, conf_list, class_list = [], [], []
    branch_num = 3
    for i in range(branch_num):
        boxes_list.append(box_process(outputs[i*3]))
        conf_list.append(outputs[i*3+2])
        class_list.append(outputs[i*3+1])

    def flatten(x):
        ch = x.shape[1]
        x = x.transpose(0,2,3,1)
        return x.reshape(-1,ch)

    boxes = np.concatenate([flatten(b) for b in boxes_list])
    box_conf = np.concatenate([flatten(c) for c in conf_list])
    class_probs = np.concatenate([flatten(c) for c in class_list])

    boxes, classes, scores, conf_keep = filter_boxes(boxes, box_conf, class_probs)
    if boxes is None:
        return None, None, None, None

    boxes[:, [0,2]] -= dx
    boxes[:, [1,3]] -= dy
    boxes /= scale
    boxes = boxes.clip(min=0)

    # 将 objectness 置信度放大 255
    scores = 1-scores
    conf_keep = conf_keep * 255
    return boxes, classes, scores, conf_keep

# ====================== 单张图片推理 ======================
def detect_single_image(img_path):
    rknn = RKNNLite(verbose=False)
    rknn.load_rknn(MODEL_PATH)
    rknn.init_runtime()

    img_name = os.path.basename(img_path)
    img = cv2.imread(img_path)
    if img is None:
        raise FileNotFoundError(f"图片无法读取: {img_path}")

    img_resized, scale, dx, dy = letterbox_resize(img, IMG_SIZE)
    input_data = np.expand_dims(img_resized, 0)
    outputs = rknn.inference(inputs=[input_data])
    boxes, classes, scores, conf_keep = post_process(outputs, scale, dx, dy)

    if boxes is not None:
        for i, box in enumerate(boxes):
            x1, y1, x2, y2 = box.astype(int)
            cls_id = classes[i]
            score = scores[i]
            cv2.rectangle(img, (x1, y1), (x2, y2), (0, 255, 0), 2)
            cv2.putText(img,
                        f"{CLASS_NAME[cls_id]}:{score:.1f}",
                        (x1, max(y1-5,0)),
                        cv2.FONT_HERSHEY_SIMPLEX,
                        0.6,
                        (0, 255, 0),
                        2)

    # 保存图像
    if conf_keep is not None and len(conf_keep) > 0:
        score_strs = ["{:.0f}".format(s) for s in conf_keep]
        name_root, ext = os.path.splitext(img_name)
        new_name = name_root + "_conf_" + "_".join(score_strs) + ext
    else:
        new_name = img_name

    save_path = os.path.join(OUTPUT_DIR, new_name)
    cv2.imwrite(save_path, img)
    print(f"{img_name} 推理完成，结果保存到: {save_path}")

    rknn.release()

# ====================== 调用 ======================
detect_single_image(IMG_PATH)