An error occurred when creating a pseudo label

[Wang-pengfei]

==> Create pseudo labels for unlabeled target domain with self-paced policy
Computing jaccard distance...
Traceback (most recent call last):
File "examples/spcl_train_uda.py", line 341, in
main()
File "examples/spcl_train_uda.py", line 109, in main
main_worker(args)
File "examples/spcl_train_uda.py", line 175, in main_worker
rerank_dist = compute_jaccard_distance(target_features, k1=args.k1, k2=args.k2)
File "/home/amax/SpCL/spcl/utils/faiss_rerank.py", line 41, in compute_jaccard_distance
_, initial_rank = search_raw_array_pytorch(res, target_features, target_features, k1)
File "/home/amax/SpCL/spcl/utils/faiss_utils.py", line 82, in search_raw_array_pytorch
I_ptr = swig_ptr_from_LongTensor(I)
File "/home/amax/SpCL/spcl/utils/faiss_utils.py", line 15, in swig_ptr_from_LongTensor
return faiss.cast_integer_to_long_ptr(
AttributeError: module 'faiss' has no attribute 'cast_integer_to_long_ptr'

[Wang-pengfei]

conda install faiss-gpu cudatoolkit=9.0 -c pytorch

I have solved this problem.

[Wang-pengfei]

==> Create pseudo labels for unlabeled target domain with self-paced policy
Computing jaccard distance...
Traceback (most recent call last):
File "examples/spcl_train_uda.py", line 341, in
main()
File "examples/spcl_train_uda.py", line 109, in main
main_worker(args)
File "examples/spcl_train_uda.py", line 175, in main_worker
rerank_dist = compute_jaccard_distance(target_features, k1=args.k1, k2=args.k2)
File "/home/amax/SpCL/spcl/utils/faiss_rerank.py", line 41, in compute_jaccard_distance
_, initial_rank = search_raw_array_pytorch(res, target_features, target_features, k1)
File "/home/amax/SpCL/spcl/utils/faiss_utils.py", line 82, in search_raw_array_pytorch
I_ptr = swig_ptr_from_LongTensor(I)
File "/home/amax/SpCL/spcl/utils/faiss_utils.py", line 15, in swig_ptr_from_LongTensor
return faiss.cast_integer_to_long_ptr(
AttributeError: module 'faiss' has no attribute 'cast_integer_to_long_ptr'

我换了一台机器，这次cuda版本是10.1，仍然出现了这个错误。
使用conda install faiss-gpu cudatoolkit=10.1 -c pytorch 依然没有解决。

[yxgeee]

听别人反映，应该是faiss的版本问题，你按照faiss 1.6.3试试

[Wang-pengfei]

之前报错的时候我安装的就是faiss 1.6.3。我这边把cuda换成9.0就可以跑通了，不知道是不是faiss和cuda版本之间有什么不兼容。

[longma2020]

Following the guide of README, I met the same problem. After checking the code in https://github.com/facebookresearch/faiss/blob/master/contrib/torch_utils.py, my solution is replacing cast_integer_to_long_ptr by cast_integer_to_idx_t_ptr for the use of faiss 1.6.5.