[찍먹분투기] YOLACT Segmentation

나온지는 조금 된 세그멘테이션을 위한 모델인 YOLACT(You Only Loot At CoefficienTs)를 돌려 본 결과를 기록해 봄. 아마 그 사이에 업데이트가 있었을테니 아래 내용이 일부 안 맞을 수도 있음.

https://github.com/dbolya/yolact

GitHub - dbolya/yolact: A simple, fully convolutional model for real-time instance segmentation.

간단 설명

• A simple, fully convolutional model for real-time instance segmentation
• DCNv2 를 이용
  • 소스코드에 포함된 DCNv2는 컴파일 단계에서 에러가 뜬다.
  • 다른. repository에서 구해다가 대체해야 한다.
• COCO Dataset을 이용하며, 학습시키려면 별도로 다운로드 받아야 한다.
• Pretrained model을 제공하므로 굳이 처음부터 학습시킬 필요는 없다.
  • https://github.com/dbolya/yolact#evaluation

찍먹하기

1. Download pretrained model
2. put the corresponding weights file in the ./weights directory
3. cd external;
   1. python setup.py build develop
      • 높은 확률로 에러 뜰거임.
        • external에 포함되어 있는 DCNv2 가 old version pytorch용이라 그런거임.
   2. mv DCNv2 _DCNv2
4. git clone https://github.com/lbin/DCNv2.git
5. make.sh
6. python eval.py --trained_model=weights/yolact_plus_resnet50_54_800000.pth --score_threshold=0.15 --top_k=15 --image=images/2020080212151464057_l.jpg:images/2020080212151464057_l_segmented.jpg
   • Log1 - 아래는 cuda 가 설치되어 있지 않기 때문에 발생한 문제임.

   Traceback (most recent call last):
     File "eval.py", line 2, in <module>
       from yolact import Yolact
     File "/home/whyun/workspace/yolact/yolact.py", line 22, in <module>
       torch.cuda.current_device()
     File "/home/whyun/anaconda3/envs/yolact-env/lib/python3.7/site-packages/torch/cuda/__init__.py", line 366, in current_device
       _lazy_init()
     File "/home/whyun/anaconda3/envs/yolact-env/lib/python3.7/site-packages/torch/cuda/__init__.py", line 172, in _lazy_init
       torch._C._cuda_init()
   RuntimeError: Unexpected error from cudaGetDeviceCount(). Did you run some cuda functions before calling NumCudaDevices() that might have already set an error? Error 804: forward compatibility was attempted on non supported HW
   

   • Log2 - 아래는 DCNv2 컴파일 하는 것을 빼 먹어서 그런거임.

   Config not specified. Parsed yolact_plus_resnet50_config from the file name.
   
   Loading model...Traceback (most recent call last):
     File "eval.py", line 1097, in <module>
       net = Yolact()
     File "/home/whyun/workspace/yolact/yolact.py", line 402, in __init__
       self.backbone = construct_backbone(cfg.backbone)
     File "/home/whyun/workspace/yolact/backbone.py", line 451, in construct_backbone
       backbone = cfg.type(*cfg.args)
     File "/home/whyun/workspace/yolact/backbone.py", line 83, in __init__
       self._make_layer(block, 128, layers[1], stride=2, dcn_layers=dcn_layers[1], dcn_interval=dcn_interval)
     File "/home/whyun/workspace/yolact/backbone.py", line 114, in _make_layer
       layers.append(block(self.inplanes, planes, stride, downsample, self.norm_layer, self.dilation, use_dcn=use_dcn))
     File "/home/whyun/workspace/yolact/backbone.py", line 23, in __init__
       padding=dilation, dilation=dilation, deformable_groups=1)
     File "/home/whyun/workspace/yolact/backbone.py", line 11, in DCN
       raise Exception('DCN could not be imported. If you want to use YOLACT++ models, compile DCN. Check the README for instructions.')
   Exception: DCN could not be imported. If you want to use YOLACT++ models, compile DCN. Check the README for instructions.
   

결과

• 입력이미지

• 출력이미지(segmented image)