huge_image_demo.py TypeError: list indices must be integers or slices, not tuple

[poirot0805]

I want to run huge_image_demo.py to inference a huge image using mask_r-cnn（only for hbb detection）

cmd:

python demo/huge_image_demo.py  demo/47.jpg     workdir/mask_rcnn_x101_32x4d_fpn_mstrain-poly_3x_coco.py      workdir/epoch_12.pth

the cfg file is from mmdetection.

env: torch1.6.0 + cuda10.1 + mmcv-full1.5.0 + mmdet2.20.0

but I got this：

load checkpoint from local path: workdir/epoch_12.pth
d:\programdata\anaconda3\mmdetection\mmdet\datasets\utils.py:69: UserWarning: "ImageToTensor" pipeline is replaced by "DefaultFormatBundle" for batch inference. It is recommended to manually replace it in the test data pipeline in your config file.
  'data pipeline in your config file.', UserWarning)
<class 'list'>
Traceback (most recent call last):
  File "demo/huge_image_demo.py", line 72, in <module>
    main(args)
  File "demo/huge_image_demo.py", line 65, in main
    args.merge_iou_thr)
  File "d:\programdata\anaconda3\mmrotate\mmrotate\apis\inference.py", line 88, in inference_detector_by_patches
    results, windows[:, :2], iou_thr=merge_iou_thr, device=device)
  File "d:\programdata\anaconda3\mmrotate\mmrotate\core\patch\merge_results.py", line 26, in merge_results
    dets[:, :2] += offset
TypeError: list indices must be integers or slices, not tuple

[poirot0805]

for results_pre_cls in zip(*results):
        tran_dets = []
        for dets, offset in zip(results_pre_cls, offsets):
            dets[:, :2] += offset  # dets is a list [error 1]
            tran_dets.append(dets)
        tran_dets = np.concatenate(tran_dets, axis=0) 
        # although I fixed error1 by enumerating dets ↓, error here because tran_dets is a list
        for item in dets:
             item[:,:2]+=offset

i'm not familiar with numpy, so i don't know the solution . T A T

[zytx121]

@jbwang1997 Please have a look.

[jbwang1997]

Hi @poirot0805. I solve this problem in pr #280. You can refer to this pr.

[poirot0805]

Hi @poirot0805. I solve this problem in pr #280. You can refer to this pr.

Thanks! it helps a lot!