guyuchao/pyramidcsa

The official implementation of "Pyramid constrained self-attention network for fast video salient object detection"

pyramidcsa's Issues

How should I load the database?

你好，我现在下载好了需要的数据集，直接更改config.py中的绝对路径可以加载数据吗？需要先生成dataloader.py中要求的list吗？请帮我解答一下，谢谢

Error when runing speed.py

Traceback (most recent call last):
File "speed.py", line 30, in
from Models import mobilenetv3temporal_PCSA as net
File "/media2/program/python/others/video_segment/PyramidCSA-master/Models/mobilenetv3temporal_PCSA.py", line 8, in
from Models.PCSA import T_Moduel
File "/media2/program/python/others/video_segment/PyramidCSA-master/Models/PCSA.py", line 2, in
import self_cuda_backend as _ext
ImportError: libcudart.so.9.0: cannot open shared object file: No such file or directory

How can self_cuda_backend find the right version?

请问下，在pretrain_distribute.py中，from utils.SalEval import SalEval？？

Training set

Thanks for your code and paper.
Did you merge the validation set and the training set of the DAVSOD for training?
Besides, the PyramidCSA uses the DAVIS as the training set while test on it. Did you split the DAVIS into a train subset and a test subset?

self_cuda_backend

May you please provide "self_cuda_backend" in PCSA module? Thank you.

Whether I can train the model on one or two GPUs?

As same as the title.

Could you please provide the code of Figure 2: Visualization of the attention map produced by the non-local operation and the CSA?

Could you please provide the code of Figure 2: Visualization of the attention map produced by the non-local operation and the CSA?

dataset

hello
I want to search for training dataset and test dataset
my email is : [email protected]

GycModule

Could you please advise where is "GycModule"? Thank you.

Model build problems

你好，我这里在将PCSA_Model模块进行打包时，始终出现一些问题，无法正常使用，想问一下当初你怎么进行模块封装的。非常感谢！！！可以加个联系方式了解一下吗？QQ: 428475501

DDP Module training stuck

Hi Prof., when I was training the model, I stuck at random epochs. Have u ever met this situation and is there any suggestion for this problem?

The training set

Hello. I want to know whether you use both the training and val set of DAVSOD or only its training set to train your model. Thanks

No module named 'self_cuda_backend'

First of all, nice work! I tried to run the fine-tuning stage but I ran into the following error.

**File "/Models/PCSA.py", line 2, in
import self_cuda_backend as _ext

ModuleNotFoundError: No module named 'self_cuda_backend'**

I then tried to run the command as instructed by you" python setup.py build develop". I ran the code again and I got another error:

File "/app/Models/PCSA.py", line 2, in
import self_cuda_backend as _ext
ImportError: libcudart.so.9.0: cannot open shared object file: No such file or directory

My torch version is 1.2 and cuda version is 10.2. Any idea about what the problem it could be? Thanks!

undefined symbol: _ZN3c105ErrorC1ENS_14SourceLocationERKSs

I face this problem when I run the script 'PCSA.py'

    import self_cuda_backend as _ext
ImportError: /xxxxxxx/Models/PCSA/self_cuda_backend.cpython-36m-x86_64-linux-gnu.so: undefined symbol: _ZN3c105ErrorC1ENS_14SourceLocationERKSs

how to solve it? please help me, thank you!

ImportError: libcudart.so.9.0: cannot open shared object file: No such file or directory

After using the commands below "Build" from ReadMe, I ran speed.py and met this error.

Do you know where the problem is? Thanks a lot!

License?

Could you please specify the license of your code? Preferably MIT

About inference

Thanks for your work. I want to ask for some details during testing phase. For a given test video sequence, do you send all frames into network to produce saliency maps once for all , or just send every T (like 5) frames?