Skip to content
This repository has been archived by the owner on Mar 19, 2024. It is now read-only.

Is it possible to use VISSL for image retrieval? #160

Closed
Zumbalamambo opened this issue Jan 29, 2021 · 4 comments
Closed

Is it possible to use VISSL for image retrieval? #160

Zumbalamambo opened this issue Jan 29, 2021 · 4 comments
Assignees
@prigoyal @iseessel
Labels
benchmark config new-request

Comments

@Zumbalamambo
Copy link

Is it possible to use VISSL for image retrieval? Please add an example to test a pretrained model
@prigoyal
Copy link
Contributor

Hi @Zumbalamambo , yes we support instance retrieval on Paris, Oxford (and revisited versions of the datasets). https://github.com/facebookresearch/vissl/blob/master/tools/instance_retrieval_test.py

we will provide an example yaml config file. Thank you for reaching out about this! :)

@prigoyal prigoyal self-assigned this Jan 29, 2021
@Zumbalamambo
Copy link
Author

@prigoyal may I know if there any template for custom dataset that can be used for image retrieval?

@iseessel iseessel self-assigned this Jun 28, 2021
@iseessel
Copy link
Contributor

Hi there is no such template, but should be able to follow the same pattern as:

vissl/vissl/utils/instance_retrieval_utils/data_util.py

Line 282 in a8622d7

class RevisitedInstanceRetrievalDataset:

facebook-github-bot pushed a commit that referenced this issue Jul 7, 2021
@QuentinDuval @facebook-github-bot
Provide an API to load / sample / stream the extracted features (#160)
243dd92 
Summary:
Propose an API to load the saved features in fairinternal/ssl_scaling#159.

The goal is to offer several ways to load extracted features:
1. merge them all in memory (all features of the dataset)
2. stream them chunk by chunk (limiting the amount of features residing in memory)
3. sample them without having to load them all (reservoir sub-sampling of features)

The (1) is the legacy option, while the (2) and (3) will allow us to compute centroids on a sub-sample of a huge dataset and then assign each sample of the huge dataset without having to load it at once.

This is therefore scaling-related work.

A future PR with the updated ClusterFit algorithm will come soon.

Pull Request resolved: fairinternal/ssl_scaling#160

Reviewed By: prigoyal

Differential Revision: D29457290

Pulled By: QuentinDuval

fbshipit-source-id: e31604ca4fdb17c7344d1181f8938f40b1aa305b
@iseessel
Copy link
Contributor

Hi there, we have open sourced image retrieval configs: #394

Let me know if you have any questions!

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Assignees

@prigoyal prigoyal

@iseessel iseessel

Labels
benchmark config new-request
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
3 participants
@prigoyal @Zumbalamambo @iseessel