Using FP16,YoloX,softnms got RuntimeError: expected scalar type Float but found Half

[Klawens]

File "/home/lsc/mmdet-2.20/mmdet/models/detectors/base.py", line 174, in forward
return self.forward_test(img, img_metas, **kwargs)
File "/home/lsc/mmdet-2.20/mmdet/models/detectors/base.py", line 147, in forward_test
return self.simple_test(imgs[0], img_metas[0], **kwargs)
File "/home/lsc/mmdet-2.20/mmdet/models/detectors/single_stage.py", line 102, in simple_test
results_list = self.bbox_head.simple_test(
File "/home/lsc/mmdet-2.20/mmdet/models/dense_heads/base_dense_head.py", line 360, in simple_test
return self.simple_test_bboxes(feats, img_metas, rescale=rescale)
File "/home/lsc/mmdet-2.20/mmdet/models/dense_heads/dense_test_mixins.py", line 37, in simple_test_bboxes
results_list = self.get_bboxes(
File "/home/lsc/mmdet-2.20/mmdet/models/dense_heads/yolox_head.py", line 294, in get_bboxes
self._bboxes_nms(cls_scores, bboxes, score_factor, cfg))
File "/home/lsc/mmdet-2.20/mmdet/models/dense_heads/yolox_head.py", line 321, in bboxes_nms
dets, keep = batched_nms(bboxes, scores, labels, cfg.nms)
File "/home/lsc/anaconda3/envs/mmdet/lib/python3.8/site-packages/mmcv/ops/nms.py", line 340, in batched_nms
dets, keep = nms_op(boxes_for_nms[mask], scores[mask], **nms_cfg)
File "/home/lsc/anaconda3/envs/mmdet/lib/python3.8/site-packages/mmcv/utils/misc.py", line 340, in new_func
output = old_func(*args, **kwargs)
File "/home/lsc/anaconda3/envs/mmdet/lib/python3.8/site-packages/mmcv/ops/nms.py", line 245, in soft_nms
dets, inds = SoftNMSop.apply(boxes.cpu(), scores.cpu(),
File "/home/lsc/anaconda3/envs/mmdet/lib/python3.8/site-packages/mmcv/ops/nms.py", line 88, in forward
inds = ext_module.softnms(
RuntimeError: expected scalar type Float but found Half

[RangiLyu]

Thanks for your report! We will check it.

[vanche]

Hi, do you have any updates? I have the same issue.

[chhluo]

Could you please provide the environment information, pkg information, and command to reproduce this error ?

[alaa-shubbak]

i still did not get the answer , i face the same issue while training my model on centriapetal_net,
with fp16 = dict(loss_scale='dynamic')

any solution for that?

[Klawens]

I solve the problem by not using the YOLOX in mmdet, btw it's not mmdet's fault

[alaa-shubbak]

ng the YOLOX in mmdet, btw it's not mmdet's fault

but i am not using yolox as a training model , i used centriapetalNet

[jbwang1997]

Hello @alaa-shubbak. Thanks for your error report. This error should be fixed after mmdet 2.23. Could you check the version of mmdet you used?

Make sure that your mmdet version is greater than or equal to 2.23.0. If the error still occurs, you can open a new issue and provide the error report, config, and environment in the new issue. We will fix the error as soon as possible.