[Bug fix] fixed fp16 inference

[xvjiarui]

No description provided.

[codecov]

Codecov Report

Merging #497 (894dbb9) into master (1483720) will decrease coverage by 0.03%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #497      +/-   ##
==========================================
- Coverage   86.51%   86.47%   -0.04%     
==========================================
  Files          97       98       +1     
  Lines        4967     5133     +166     
  Branches      806      829      +23     
==========================================
+ Hits         4297     4439     +142     
- Misses        517      535      +18     
- Partials      153      159       +6     

Flag	Coverage Δ
unittests	86.47% <ø> (-0.04%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
mmseg/datasets/pascal_context.py	76.47% <0.00%> (-3.53%)	⬇️
mmseg/datasets/pipelines/transforms.py	97.12% <0.00%> (-0.87%)	⬇️
mmseg/models/backbones/__init__.py	100.00% <0.00%> (ø)
mmseg/models/backbones/vit.py	87.97% <0.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 1483720...894dbb9. Read the comment docs.

[clownrat6]

Re-Benchmark

The benchmark testing is conducted on NVIDIA® Tesla® V100.

Pytorch 1.8 + Cuda10.1

Cityscapes

Method	Backbone	Crop Size	Lr schd	Mem (GB)	Inf time (fps)	mIoU	mIoU(ms+flip)
FCN	R-101-D8	512x1024	80000	5.37	8.64	77.32	78.46
PSPNet	R-101-D8	512x1024	80000	5.34	8.77	79.22	80.33
DeepLabV3	R-101-D8	512x1024	80000	5.75	3.86	80.73	81.82
DeepLabV3+	R-101-D8	512x1024	80000	6.35	7.87	80.29	80.98

Pytorch 1.5 + Cuda10.1

Cityscapes

Method	Backbone	Crop Size	Lr schd	Mem (GB)	Inf time (fps)	mIoU	mIoU(ms+flip)
FCN	R-101-D8	512x1024	80000	5.50	8.26	78.02	78.89
PSPNet	R-101-D8	512x1024	80000	5.47	8.69	79.28	80.62
DeepLabV3	R-101-D8	512x1024	80000	5.91	3.66	79.97	81.04
DeepLabV3+	R-101-D8	512x1024	80000	6.46	7.48	79.96	80.89

[clownrat6]

Pytorch 1.3 + Cuda10.1

When I run tools/benchmark.py [config_path] [ckpt_path], there comes an error:

Traceback (most recent call last):                                                                                                                
  File "tools/benchmark.py", line 85, in <module>                                                                                                                                                           
    main()                                                                                                                                                                                                  
  File "tools/benchmark.py", line 66, in main                                                                                                                                                               
    model(return_loss=False, rescale=True, **data)                                                                                                                                                          
  File "/mnt/cache/chengzesen/anaconda3/envs/pt1.3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 541, in __call__                                                                           
    result = self.forward(*input, **kwargs)                                                                                                                                                                 
  File "/mnt/cache/chengzesen/workspace/mmcv/mmcv/parallel/data_parallel.py", line 42, in forward                                                                                                           
    return super().forward(*inputs, **kwargs)                                                                                                                                                               
  File "/mnt/cache/chengzesen/anaconda3/envs/pt1.3/lib/python3.6/site-packages/torch/nn/parallel/data_parallel.py", line 150, in forward                                                                    
    return self.module(*inputs[0], **kwargs[0])                                                                                                                                                             
  File "/mnt/cache/chengzesen/anaconda3/envs/pt1.3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 541, in __call__                                                                           
    result = self.forward(*input, **kwargs)                                                                                                                                                                 
  File "/mnt/cache/chengzesen/workspace/mmcv/mmcv/runner/fp16_utils.py", line 124, in new_func                                                                                                              
    output = old_func(*new_args, **new_kwargs)                                                                                                                                                              
  File "/mnt/cache/chengzesen/workspace/mmsegmentation/mmseg/models/segmentors/base.py", line 124, in forward                                                                                               
    return self.forward_test(img, img_metas, **kwargs)                                                                                                                                                      
  File "/mnt/cache/chengzesen/workspace/mmsegmentation/mmseg/models/segmentors/base.py", line 106, in forward_test                                                                                          
    return self.simple_test(imgs[0], img_metas[0], **kwargs)                                                                                                                                                
  File "/mnt/cache/chengzesen/workspace/mmsegmentation/mmseg/models/segmentors/encoder_decoder.py", line 271, in simple_test                                                                                
    seg_pred = seg_logit.argmax(dim=1)                                                                                                                                                                      
RuntimeError: "argmax_cuda" not implemented for 'Half'