train about cnnimageretrieval-pytorch

[Salehbigdeli]

The default dataset for training is retrieval-SfM-120k it's not necessary to set it for train. Use newest version of code.

[filipradenovic]

Have you tried running the train example from README training section? I.e., have you tried running the following line:

python3 -m cirtorch.examples.train YOUR_EXPORT_DIR --gpu-id '0' --training-dataset 'retrieval-SfM-120k' 
            --test-datasets 'roxford5k,rparis6k' --arch 'resnet101' --pool 'gem' --loss 'contrastive' 
            --loss-margin 0.85 --optimizer 'adam' --lr 1e-6 --neg-num 5 --query-size=2000 
            --pool-size=22000 --batch-size 5 --image-size 362

[Garamda]

I eliminated quotation marks ex) 'retrieval-SfM-120k' -> retrieval-SfM-120k and it worked. I don't know why the quotation marks need to be removed though. Other quotation marks in the command should be removed in the same way.

p.s. I'm using Window 10, not Linux.

[filipradenovic]

@Garamda

p.s. I'm using Window 10, not Linux.

This might be the reason, we have only tested it on Linux, so far. It is good to know the code runs on Windows as well.

[Garamda]

@filipradenovic

However, because of Linux commands like wget or rm, I couldn't go further. I think I should use Linux.

[filipradenovic]

@Garamda

If the problem is only in the download script, then this can be solved by manually downloading necessary files from cnnimageretrieval website. It should be clear from the download script how the data should be arranged (folder structure). I don't know if it is too much work to do it manually though, but it is only done once, so it might be worth it.

Hope this helps anyhow.

[Garamda]

@filipradenovic

That is great idea. Thank you :)

[filipradenovic]

@Garamda if you do manage to run it with Windows, can you post your findings here? It might be very useful for many other Windows users, and we would be very grateful. If it is easier for you, you can even open a new issue named 'Running the toolbox on Windows'. Thank you.

[Garamda]

@filipradenovic

Alright. If I can, I ll do it :)