Closed LearningJack closed 2 years ago
You need to use webdataset when downloading with img2dataset
On Mon, May 23, 2022, 14:54 LearningJack @.***> wrote:
[image: issue] https://user-images.githubusercontent.com/38984606/169823839-816f20ec-ff59-4069-b655-9fa707ff2545.jpg Could you help me? When I run the clip-retrieval and the metadata shows containing folders name and files name after run reorder-embeddings example-key --metadata-folder ./img_emb_test/metadata, how can I save the metadata without containing folders name?
— Reply to this email directly, view it on GitHub https://github.com/rom1504/clip-retrieval/issues/149, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAR437W3R22YQJUIZQJI5BTVLN5ZVANCNFSM5WVYGRTQ . You are receiving this because you are subscribed to this thread.Message ID: @.***>
When I use the webdataset format tar, I run clip-retrieval inference --input_dataset "{000000}.tar" --output_folder ../img_emb_test --input_format webdataset, but it's wrong. Could you tell me what is wrong about the commands I run and then how I should run the correct command? Thank you for your reply and help.
Looks like you don't have any file called {000000}.tar Maybe you mean 000000.tar or {000000..000005}.tar ?
Could you help me? When I run the clip-retrieval and the metadata shows containing folders name and files name after run reorder-embeddings example-key --metadata-folder ./img_emb_test/metadata, how can I save the metadata without containing folders name?