Datasets:

friedrichor
/

MSR-VTT

Tasks:

Video Classification

Modalities:

Formats:

Languages:

Size:

Libraries:

Dataset card Data Studio Files Files and versions Community

friedrichor commited on 4 days ago

Commit

fabb858

·

verified ·

1 Parent(s): a87b463

Update README.md

Files changed (1) hide show

README.md +2 -1

README.md CHANGED Viewed

@@ -8,7 +8,7 @@ configs:
   data_files:
   - split: train
     path: "msrvtt_train_7k.json"
-- config_name: test_1k, The current mainstream split method.
   data_files:
   - split: test
     path: "msrvtt_test_1k.json"
@@ -22,6 +22,7 @@ size_categories:
 - 1K<n<10K
 ---
 [MSRVTT](https://openaccess.thecvf.com/content_cvpr_2016/html/Xu_MSR-VTT_A_Large_CVPR_2016_paper.html) contains 10K video clips and 200K captions.
 We adopt the standard `1K-A split` protocol, which was introduced in [JSFusion](https://openaccess.thecvf.com/content_ECCV_2018/html/Youngjae_Yu_A_Joint_Sequence_ECCV_2018_paper.html) and has since become the de facto benchmark split in the `Text-Video Retrieval` field.

   data_files:
   - split: train
     path: "msrvtt_train_7k.json"
+- config_name: test_1k
   data_files:
   - split: test
     path: "msrvtt_test_1k.json"
 - 1K<n<10K
 ---
 [MSRVTT](https://openaccess.thecvf.com/content_cvpr_2016/html/Xu_MSR-VTT_A_Large_CVPR_2016_paper.html) contains 10K video clips and 200K captions.
 We adopt the standard `1K-A split` protocol, which was introduced in [JSFusion](https://openaccess.thecvf.com/content_ECCV_2018/html/Youngjae_Yu_A_Joint_Sequence_ECCV_2018_paper.html) and has since become the de facto benchmark split in the `Text-Video Retrieval` field.