tensors is empty when use part_base_engine _feature_extraction

[Đặng Lê Minh]

So I had this error :
TimeoutError:
[WinError 10060] A connection attempt failed because the connected party did not properly respond after a period of time, or established connection failed because connected host has failed to respond

when trying to run this script :"python scripts/main.py --config-file configs/bpbreid/bpbreid_market1501_test.yaml". And I tried to fix it by downloading the dataset from drive and set the directories as in your instruction

After that, everything runs just fine until it comes to _feature_extraction at part_based_engine.py and I'm not sure if data_loader is null or not(I'm totally new to Python) but the for loop isn't running at all.
"RuntimeError: torch.cat(): expected a non-empty list of Tensors".
I would appreciate if you could guide me a little bit.

[Vladimir Somers]

Hi, are all you directory non empty? I guess this happen because the "data_loader" in "_feature_extraction" is empty (can you verify that?). Can you check in "torchreid/data/datasets/image/market1501.py", where the dataset is loaded, if "train", "query" and "gallery" in "init" are non empty?

[Đặng Lê Minh]

Yes all the directories are non empty
.
But apparently the train query and gallery is empty

[Vladimir Somers]

Then there's an issue with you path specification, can you check if self.train_dir, self.query_dir, self.galery_dir match your directory path on disk?

[Đặng Lê Minh]

Well they match but instead of having only one "\" the dir has 2 is it wrong or it's just how path works in python

F:\ReID\bpbreid\reid-data\Market-1501-v15.09.15\bounding_box_test

[Vladimir Somers]

I'm not sure what's happening if the path is correct, I'm sorry, but at least you found the root cause for the error: now you should figure out why the dataset is not located properly (it should definitely be an issue with your path or Windows related stuff)

[Đặng Lê Minh]

It's okay thanks for helping me.

[Đặng Lê Minh]

Okay so I think I found the problem, apparently the function is trying to get files with .jpg but the dataset contains these instead .npy files, should I reformat it to jpg?

[Vladimir Somers]

You should download the dataset on the internet, and then place the masks downloaded on Google Drive in the "masks" folder. The final dataset folder structure should look like this:

Market-1501-v15.09.15
├── bounding_box_test
├── bounding_box_train
├── masks
│   └── pifpaf_maskrcnn_filtering
│       ├── bounding_box_test
│       ├── bounding_box_train
│       └── query
└── query

With images (.jpg) inside bounding_box_test/bounding_box_train/query on the first level and .npy with the masks inside pifpaf_maskrcnn_filtering, with the same subfolder structure

[Đặng Lê Minh]

I think it's a problem with file extensions, apparently it detects those files as .npy files not .jpg, i modified the code and it works

Update: It didn't work when it comes to feature extraction, I will try to find a way to change it to .jpg files
Update: I finally understand this line"You should download the dataset on the internet". Thank you so much