Update README.md

README.md

@@ -99,6 +99,20 @@ wget https://huggingface.co/datasets/WenhaoWang/VidProM/resolve/main/VidProM_uni
 
 ``VidProM_embed.hdf5`` is the 3072-dim embeddings of our prompts. They are embedded by text-embedding-3-large, which is the latest text embedding model of OpenAI.
 
+It can easily read by
+
+```
+import numpy as np
+import h5py
+def read_descriptors(filename):
+    hh = h5py.File(filename, "r")
+    descs = np.array(hh["embeddings"])
+    names = np.array(hh["uuid"][:], dtype=object).astype(str).tolist()
+    return names, descs
+
+uuid, features = read_descriptors('VidProM_embed.hdf5')
+```
+
 ``original_files`` are the HTML files collected by DiscordChatExporter.
 
 ``pika_videos``, ``vc2_videos``, ``t2vz_videos``, and ``ms_videos`` are the generated videos by 4 state-of-the-art text-to-video diffusion models. Each contains 30 tar files.