yolov8n模型在k230上推理卡死问题怎么解决

Viewed 21

问题描述


训练好的yolov8n权重文件正常转onnx后正常转kmodel,现在不想开摄像头,只想拿一个文件夹推理看精度,每次读图后均卡死,是读图的问题吗,图片分辨率为320×320,测试图片为39张,认为为识别花的五种状态并定位,因为身边没有这种花所以只能拿测试图片测试精度,但是每次在ide上跑都卡死了(模型加载没问题,读图就卡死了,终端不动了)

复现步骤


from libs.YOLO import YOLOv8
from libs.Utils import *
import os, sys, gc, time
import image

if name=="main":

kmodel_path = "/sdcard/best.kmodel"
labels = ["rose_bud","rose_little_open","rose_half_bloom","rose_full_bloom","rose_fading"]
model_input_size = [320, 320]
image_folder = "/sdcard/data"                 # 确保这个文件夹里有图片
output_folder = "/sdcard/output_results"

confidence_threshold = 0.5
nms_threshold = 0.45
# ====================================================

# 确保输出文件夹存在
try:
    os.mkdir(output_folder)
except OSError:
    pass

# 初始化 YOLOv8
yolo = YOLOv8(task_type="detect", mode="image", kmodel_path=kmodel_path, labels=labels,
              rgb888p_size=model_input_size, model_input_size=model_input_size,
              conf_thresh=confidence_threshold, nms_thresh=nms_threshold,
              max_boxes_num=50, debug_mode=1)
yolo.config_preprocess()

# 获取文件夹里所有的图片文件
img_list = [f for f in os.listdir(image_folder) if f.endswith('.jpg') or f.endswith('.jpeg') or f.endswith('.png')]
total_images = len(img_list)
print(f"找到 {total_images} 张图片,开始测试...")

total_time = 0

for img_name in img_list:
    os.exitpoint() # 防止跑太久被系统看门狗强行打断
    img_path = image_folder + "/" + img_name

    try:
        # 读取图片
        img = image.Image(img_path)
        if not img:
            print(f"无法读取图片: {img_name}")
            continue

        # --- 开始计时 (使用 ticks_ms) ---
        start_time = time.ticks_ms()

        # 推理
        res = yolo.run(img)

        # --- 结束计时 ---
        end_time = time.ticks_ms()
        inference_time = time.ticks_diff(end_time, start_time)
        total_time += inference_time

        print(f"[{img_name}] 耗时: {inference_time} ms, 检出目标数: {len(res)}")

        # 如果检测到目标,打印坐标和类别
        if len(res) > 0:
            for obj in res:
                print(f"  -> {obj}") # 这里会打印出类似 [x, y, w, h, class_id, score]

        # 💡 重点:把框画在图上,并保存到 output_results
        # 这样你可以把这些图片拷回电脑,肉眼对比精度!
        if len(res) > 0:
            yolo.draw_result(res, img)

        save_path = output_folder + "/" + img_name
        img.save(save_path)

        # 释放内存
        del img
        gc.collect()

    except Exception as e:
        print(f"处理 {img_name} 时出错: {e}")

# 算总账
if total_images > 0:
    avg_time = total_time / total_images
    fps = 1000 / avg_time
    print("====================================")
    print(f"测试完成!共处理 {total_images} 张图片。")
    print(f"平均每张耗时(含前处理+推理+后处理): {avg_time:.1f} ms")
    print(f"等效 FPS: {fps:.1f}")
    print("====================================")
    print("检测结果图已保存到 /sdcard/output_results,请拷贝回电脑查看精度。")

yolo.deinit()

硬件板卡


庐山派

其他信息


image.png就这个状态

1 Answers

YOLO读入的是ulab.numpy.ndarray格式的数据,并且需要用chw的数据,你可以参考examples/20-YOLO-Module-Examples的读图方式测试,在libs.Utils中提供了read_image方法:

from libs.Utils import *

img_path="写你自己的图片路径"
img,img_ori=read_image(img_path)
...
res=yolo.run(img)