test online model about trackit

[JudasDie]

Thanks for your great work!
I want to know how to test with online model like OceanV18on.pth?
I have tried python tracking/test_ocean.py --arch Ocean --resume snapshot/OceanV18on.pth --dataset VOT2018 --online True but didn't work.

Hi, could you provide a detailed screenshot?

[zzzmm1]

===> init Siamese <====
load pretrained model from snapshot/OceanV18on.pth
remove prefix 'module.'
remove prefix 'feature_extractor.'
missing keys:[]
unused checkpoint keys:['classifier.filter_optimizer.log_step_length', 'classifier.feature_extractor.0.weight', 'classifier.filter_initializer.filter_conv.weight', 'classifier.filter_optimizer.label_map_predictor.weight', 'classifier.filter_initializer.filter_conv.bias', 'classifier.filter_optimizer.filter_reg', 'classifier.filter_optimizer.spatial_weight_predictor.weight', 'classifier.filter_optimizer.target_mask_predictor.0.weight']
====> warm up <====
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 100/100 [00:01<00:00, 50.45it/s]

and then finished without test results

[JudasDie]

===> init Siamese <====
load pretrained model from snapshot/OceanV18on.pth
remove prefix 'module.'
remove prefix 'feature_extractor.'
missing keys:[]
unused checkpoint keys:['classifier.filter_optimizer.log_step_length', 'classifier.feature_extractor.0.weight', 'classifier.filter_initializer.filter_conv.weight', 'classifier.filter_optimizer.label_map_predictor.weight', 'classifier.filter_initializer.filter_conv.bias', 'classifier.filter_optimizer.filter_reg', 'classifier.filter_optimizer.spatial_weight_predictor.weight', 'classifier.filter_optimizer.target_mask_predictor.0.weight']
====> warm up <====
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 100/100 [00:01<00:00, 50.45it/s]

and then finished without test results

Hi, the message has been well-received. I would give you feedback asap 😄.

[JudasDie]

===> init Siamese <====
load pretrained model from snapshot/OceanV18on.pth
remove prefix 'module.'
remove prefix 'feature_extractor.'
missing keys:[]
unused checkpoint keys:['classifier.filter_optimizer.log_step_length', 'classifier.feature_extractor.0.weight', 'classifier.filter_initializer.filter_conv.weight', 'classifier.filter_optimizer.label_map_predictor.weight', 'classifier.filter_initializer.filter_conv.bias', 'classifier.filter_optimizer.filter_reg', 'classifier.filter_optimizer.spatial_weight_predictor.weight', 'classifier.filter_optimizer.target_mask_predictor.0.weight']
====> warm up <====
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 100/100 [00:01<00:00, 50.45it/s]

and then finished without test results

Hi, I think I have found the issue. Before running online, please check your results directory. If you run offline model before, it will generate a directory, namely result/VOT2018/Ocean. Please remove this directory.

[zzzmm1]

Ok, thanks for your help !

[zzzmm1]

Hi, Zhipeng.
I met another problem "RuntimeError: Error building extension '_prroi_pooling'".
Is there any pre requirements？

[JudasDie]

Hi, Zhipeng.
I met another problem "RuntimeError: Error building extension '_prroi_pooling'".
Is there any pre requirements？

Yes, the code will automatically compile precise ROI pooling. If you fail to compile it, please check the python environment and CUDA.

[zzzmm1]

Hi, Zhipeng.
I met another problem "RuntimeError: Error building extension '_prroi_pooling'".
Is there any pre requirements？

Yes, the code will automatically compile precise ROI pooling. If you fail to compile it, please check the python environment and CUDA.

My environment was built as the install.sh, pytorch 1.1.0 with cuda 10.0, but still failed.
And I want to know which part in the online model will use precise ROI pooling?
Could I pre build precise ROI pooling first and then test online , will it makes any difference?

[JudasDie]

Hi, Zhipeng.
I met another problem "RuntimeError: Error building extension '_prroi_pooling'".
Is there any pre requirements？

Yes, the code will automatically compile precise ROI pooling. If you fail to compile it, please check the python environment and CUDA.

My environment was built as the install.sh, pytorch 1.1.0 with cuda 10.0, but still failed.
And I want to know which part in the online model will use precise ROI pooling?
Could I pre build precise ROI pooling first and then test online , will it makes any difference?

The online model use ROI pooling to extract target features, please read code in lib/models/online or refer to this. As for the compiling, I recommend you to the ROI pooling repo to find more information on compiling errors.

[zzzmm1]

I found the problem is caused by the downloading zip file in the lib/models/online/external/PreciseRoIPooling instead of using git clone PreciseRoIPooling. It breaks the symbolic links.

[n1-k0]

@zzzmm1 Hi! I wonder how to solve this problem. I use git clone PreciseRoIPooling and put it into the lib/models/online/external/PreciseRoIPooling
but still have this error