inference_detector with multiple images?

Is it possible to call inference_detector (mmdet/apis/inference.py) with multiple images (list of image-paths)? The doku says yes, but calling the function with a list of image paths throws an exception. Does anyone have any idea what to change, so that multiple images are possible?

# build the data pipeline
test_pipeline = [LoadImage()] + cfg.data.test.pipeline[1:]
test_pipeline = Compose(test_pipeline)
# prepare data
data = dict(img=img)
data = test_pipeline(data)
data = scatter(collate([data], samples_per_gpu=1), [device])[0]

For a Python newbie, these few lines of code are very confusing.

[minan19605]

my code for inference multiple images in one directory, just for your reference:

from mmdet.apis import init_detector, inference_detector, show_result_pyplot import mmcv #import matplotlib.pyplot as plt import cv2 import glob import os import sys import time import argparse

‘’’ python tools/my_inference.py
–cfg configs/my_config/htc_without_semantic_r50_fpn_1x_coco_linux.py
–output-dir work_dirs/htc/20200608/infer_paper/
–wts work_dirs/htc/20200608/epoch_200.pth
…/coco/infer_img_dir ‘’’

def parse_args(): parser = argparse.ArgumentParser(description=‘End-to-end inference’) parser.add_argument( ‘–cfg’, dest=‘cfg’, help=‘cfg model file (/path/to/model_config.yaml)’, default=None, type=str ) parser.add_argument( ‘–wts’, dest=‘weights’, help=‘weights model file (/path/to/model_weights.pkl)’, default=None, type=str ) parser.add_argument( ‘–output-dir’, dest=‘output_dir’, help=‘directory for visualization pdfs (default: /tmp/infer_simple)’, default=‘/tmp/infer_simple’, type=str ) parser.add_argument( ‘–image-ext’, dest=‘image_ext’, help=‘image file name extension (default: jpg)’, default=‘jpg’, type=str ) parser.add_argument( ‘–always-out’, dest=‘out_when_no_box’, help=‘output image even when no object is found’, action=‘store_true’ ) parser.add_argument( ‘–output-ext’, dest=‘output_ext’, help=‘output image file format (default: jpg)’, default=‘jpg’, type=str ) parser.add_argument( ‘–thresh’, dest=‘thresh’, help=‘Threshold for visualizing detections’, default=0.3, type=float ) parser.add_argument( ‘im_or_folder’, help=‘image or folder of images’, default=None ) if len(sys.argv) == 1: parser.print_help() sys.exit(1) return parser.parse_args()

def main(args):

config_file = args.cfg
checkpoint_file = args.weights
thresh=args.thresh
model = init_detector(config_file, checkpoint_file, device='cuda:0')
#model = init_detector(config_file, checkpoint_file, device='cpu')

if os.path.isdir(args.im_or_folder):
    im_list = glob.iglob(args.im_or_folder + '/*.' + args.image_ext)
else:
    im_list = [args.im_or_folder]

total_t = time.time()
for i, im_name in enumerate(im_list):
    print('Get img {}'.format(im_name))
    one_t = time.time()
    if(args.image_ext == args.output_ext):
        output_name = os.path.join(
            args.output_dir, '{}'.format(os.path.basename(im_name))
        )
    else:
        output_name = os.path.join(
            args.output_dir, '{}'.format(os.path.basename(im_name) + '.' + args.output_ext)
        )
    
    result = inference_detector(model, im_name)
    if hasattr(model, 'module'):
        model = model.module
    pred_img = model.show_result(im_name, result, score_thr=thresh, show=False)
    cv2.imwrite(output_name,pred_img)
    print('write {}'.format(output_name))
    print('One image time {:.3f}s'.format(time.time()-one_t))
print("Total inference time is {:.3f}s".format(time.time() - total_t))

if name == ‘main’: args = parse_args() main(args)

[leemengwei]

@Kaeseknacker Actually it won’t be “much” faster. We have tried forward the network with multiple images but it is only slightly faster under some circumstance (e.g., input images have the same resolution), and even slower sometimes.

Why is that batch inference only slightly faster instead of multiple times faster? It’s because of: GPU hardware/cuda version/python limitation/pytorch or mmdetection implementation? Could you share more insights? Thanks!