mmdetection中使用训练好的模型批量图片推理并保存到文件夹

# Copyright (c) OpenMMLab. All rights reserved.
import asyncio
import numpy as np
from argparse import ArgumentParser

from mmdet.apis import (async_inference_detector, inference_detector,
                        init_detector, show_result_pyplot)

import mmcv
import os

import warnings
warnings.filterwarnings("ignore") 


def get_img_file(file_name):
    imagelist = []
    for parent, dirnames, filenames in os.walk(file_name):
        for filename in filenames:
            if filename.lower().endswith(('.bmp', '.dib', '.png', '.jpg', '.jpeg', '.pbm', '.pgm', '.ppm', '.tif', '.tiff')):
                imagelist.append(os.path.join(parent, filename))
        
        return imagelist

def parse_args():
    parser = ArgumentParser()
    parser.add_argument('--img_dir', default='/home/bhts/MMDetection/mmdetection/demo/1/',help='Image file')
    parser.add_argument('--out_dir', default='./output/',help='Image file')
    parser.add_argument('--config',default='../configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py', help='Config file')
    parser.add_argument('--checkpoint',default='../checkpoints/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth', help='Checkpoint file')
    parser.add_argument(
        '--device', default='cpu', help='Device used for inference') #cuda:0
    parser.add_argument(
        '--score-thr', type=float, default=0.9, help='bbox score threshold')
    parser.add_argument(
        '--async-test',
        action='store_true',
        help='whether to set async options for async inference.')
    args = parser.parse_args()
    return args


def main(args):
    
    model = init_detector(args.config, args.checkpoint, device=args.device)

    out_dir = args.out_dir
    if not os.path.exists(out_dir):  #判断是否存在文件夹如果不存在则创建为文件夹
       os.makedirs(out_dir)

    images = get_img_file(args.img_dir)
    
    for image in images:
      print(image)
	 # 测试单张图片并展示结果
      img = mmcv.imread(image)    # 或者 ，这样图片仅会被读一次 img = 'demo.jpg'
      result = inference_detector(model, img)
    
      #bboxes_scores = np.vstack(result)
      #bboxes=bboxes_scores[:,:4]
      #score=bboxes_scores[:,4]
      #labels = [
      #           np.full(bbox.shape[0], i, dtype=np.int32)
      #           for i, bbox in enumerate(result)
      #       ]
      #labels = np.concatenate(labels)
      #print(bboxes_scores)
      #print(labels)
      #print(result)
      out_file = out_dir + image.split('/')[-1]
      model.show_result(img,result, score_thr=args.score_thr,out_file=out_file)
    




if __name__ == '__main__':
    args = parse_args()
    if args.async_test:
        asyncio.run(async_main(args))
    else:
        main(args)

版权声明：本文为CSDN博主「平凡中寻找不平凡」的原创文章，遵循CC 4.0 BY-SA版权协议，转载请附上原文出处链接及本声明。
原文链接：https://blog.csdn.net/weixin_43804210/article/details/122110492