Open haozidede opened 2 years ago
The lmdb file is one I adopted from another repo, as the original Clip4clip consumes lots of time reading the data. The usage of lmdb was in tuning and currently, I have no plan to continue. However, this can be eliminated by using an efficient decoding package, i.e., decord. I have committed this modification already and hope it may help you.
Of course, a reasonable setup of num_worker may also help.
IN'/dataloaders/dataloader_msrvtt_retrieval.py', do MSRVTT_TrainDataLoader.init
What is ' self.env = lmdb.open( '/projects/np28/mfhan/ClipBERT/vis_db/msrvtt', readonly=True, create=False) # readahead=not _check_distributed() ' used for? And the mentioned file does not exists?
Could you please help me solve the problem?