minghanli / stmask Goto Github PK

Thank you for the great project!!
I try to test your project using provided weights.
However, it seems that the link for the provided weights is incorrect.
Could you please check the links?

NameError: name 'spatial_correlation_sample' is not defined

I urge you to upload this python package .

Thank you for sharing the repo!

We are trying to reproduce the results of Youtube2019 based on your instructions. We tried two different pretrained weights, but got the same errors:

RuntimeError: invalid spatial size of offset, expected height: %d width: %d, but got height: %d width: %d50784880 (deform_conv_shape_check at ./mmcv/ops/csrc/pytorch/deform_conv_cuda.cu:177)

The two commands we used were as follows:
python eval.py --config=STMask_plus_base_ali_config --trained_model=weights/STMask_plus_base_ali.pth --mask_det_file=weights/results.json

python eval.py --config=STMask_plus_base_ada_config --trained_model=weights/STMask_plus_base_ada.pth --mask_det_file=weights/results.json

We would really appreciate some help, thank you very much!

valid_sub_YouTube_VOS_dataset = dataset_base.copy({
    'img_prefix': '../datasets/YouTube_VOS2019/train/JPEGImages',
    'ann_file': '../datasets/YouTube_VOS2019/annotations_instances/valid_sub.json',
    'test_mode': False,
})

@MinghanLi
It sounds like you split the train jsons into train_sub and valid_sub for epoch/iterations evaluation.
Could you provide more details about this one such as how to split to reproduce your results? I could not find it across the codes.

And, is that the train-sub and valid_subare only used in developments, and all reported results are trained by full train set after 12 epochs for YT and 20 epochs for OVIS?

Thanks!

Dear author, thanks for your great job! I'm trying to train STMask and I find the batchsize and max_iter you gave in the code is different from the paper.
In the paper, you set bz 16 and total iters 160000.

But in the code you set bz 8 and max_iters 250000

So, I want to know the exact paramters to get the final result.

ImportError: /home/lin/.conda/envs/stmask/lib/python3.7/site-packages/spatial_correlation_sampler_backend.cpython-37m-x86_64-linux-gnu.so: undefined symbol: _ZNK2at10TensorBase8data_ptrIdEEPT_v

File "/STMask/layers/modules/track_to_segment_head.py", line 61, in correlate
out_corr = out_corr.view(b, ph*pw, h, w) / x1.size(1)
RuntimeError: CUDA error: invalid device function
段错误 (核心已转储)

Thank you for sharing the code!

when I run the code, I met the error above.

I would really appreciate your help to fix this problem.

Thank your very much~

I found a problem.
This code runs uncorrectly in Pytorch1.5.0
please use Pytorch 1.4.0
and when install mmcv, please use pip install mmcv-full==1.1.2 -f https://download.openmmlab.com/mmcv/dist/cu101/torch1.4.0/index.html