Can add_pca.py be used on the pitts dataset

[DonMuv]

HI,
Can the add_pca.py be used on the pitts30 dataset?Does pca_train_set use the pitts30's training set?

[Tobias-Fischer]

Hi. Please see #24 - just use the Nanne repo, the trained model can then be PCA-ed using add_pca.py. Feel free to reopen if you have more questions.

[vturrisi]

Hi,
When adding the PCA, I should use pitts30's training set instead of mapillary, right?

[Tobias-Fischer]

Hi. You can apply PCA in both cases; the training dataset does not matter.

[vturrisi]

Sure, but in:

Patch-NetVLAD/add_pca.py

Lines 125 to 129 in d2b341e

	pca_train_set = MSLS(opt.dataset_root_dir, mode='test', cities='train',
	transform=input_transform(),
	bs=int(config['train']['cachebatchsize']), threads=opt.threads,
	margin=float(config['train']['margin']),
	exclude_panos=exlude_panos_training)

the dataset is fixed as mapillary, so I need to modify the code in order to use a dataset that's compatible with Pittsburgh, right?

[Tobias-Fischer]

Ah, I see what you mean. Indeed the code needs modification at this spot.

@StephenHausler - shall we sit together to amend the code so that at least for PCA we also enable Pitts?

[vturrisi]

Would be super good!

[StephenHausler]

@Tobias-Fischer yep lets do that