From e6c06a0245fac63dceb38bc9bd4fecd033dae735 Mon Sep 17 00:00:00 2001
From: Guanzhi Wang <wangguanzhi2015@gmail.com>
Date: Sat, 1 Apr 2023 21:10:39 -0700
Subject: [PATCH] Update training data format

---
 README.md | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)

diff --git a/README.md b/README.md
index df6037c..6bbc0ce 100644
--- a/README.md
+++ b/README.md
@@ -83,6 +83,22 @@ You can also find two sample env implementations `HuntCowDenseRewardEnv` and `Co
 
 We provide [a superset of 640K video clips](https://drive.google.com/file/d/1cLXC64Cu2EJj2nsb4K0ajl8qqX6l0lKd/view?usp=sharing) we used for pre-training. You can subsample from this set and grow start/end timestamps as you like (basically what we did for training MineCLIP).
 
+Format:
+```python
+{
+    "keyword1": list[
+        [video_id1, start_timestamp1, end_timestamp1],
+        [video_id2, start_timestamp2, end_timestamp2],
+        [video_id3, start_timestamp3, end_timestamp3],
+    ],
+    "keyword2": list[
+        [video_id4, start_timestamp4, end_timestamp4],
+        ...
+    ],
+    ...
+}
+```
+
 # Paper and Citation
 
 Our paper is posted on [Arxiv](https://arxiv.org/abs/2206.08853). If you find our work useful, please consider citing us!