retinanet错误

retinanet-evaluate --gpu 1 --save-path ../converted_models/images --no-resize csv /home/robot/woodcore/code/aae/detection_utils/output/test/annotations.csv /home/robot/woodcore/code/aae/detection_utils/output/test/classes.csv ../converted_models/resnet50_csv_39_inference.h5

Traceback (most recent call last):
File "/home/robot/miniconda3/envs/retinanet/bin/retinanet-evaluate", line 8, in <module>
sys.exit(main())
File "/home/robot/miniconda3/envs/retinanet/lib/python3.6/site-packages/keras_retinanet/bin/evaluate.py", line 172, in main
save_path=args.save_path
File "/home/robot/miniconda3/envs/retinanet/lib/python3.6/site-packages/keras_retinanet/utils/eval.py", line 174, in evaluate
all_detections, all_inferences = _get_detections(generator, model, score_threshold=score_threshold, max_detections=max_detections, save_path=save_path)
File "/home/robot/miniconda3/envs/retinanet/lib/python3.6/site-packages/keras_retinanet/utils/eval.py", line 109, in _get_detections
draw_annotations(raw_image, generator.load_annotations(i), label_to_name=generator.label_to_name)
File "/home/robot/miniconda3/envs/retinanet/lib/python3.6/site-packages/keras_retinanet/utils/visualization.py", line 105, in draw_annotations
draw_caption(image, annotations['bboxes'][i], caption)
File "/home/robot/miniconda3/envs/retinanet/lib/python3.6/site-packages/keras_retinanet/utils/visualization.py", line 45, in draw_caption
cv2.putText(image, caption, (b[0], b[1] - 10), cv2.FONT_HERSHEY_PLAIN, 1, (0, 0, 0), 2)
TypeError: Layout of the output array img is incompatible with cv::Mat (step[ndims-1] != elemsize or step[1] != elemsize*nchannels)

解决办法：

把eval.py文件中第78，79两行

raw_image = generator.load_image(i)

image, scale = generator.resize_image(raw_image.copy())

改成

raw_image = generator.load_image(i)

raw_image = raw_image.copy()

image, scale = generator.resize_image(raw_image)

即：

要原始数组的副本。