igorsterner commited on
Commit
677ef11
·
verified ·
1 Parent(s): 50499fb
Files changed (7) hide show
  1. .gitattributes +1 -0
  2. README.md +94 -0
  3. keys.zip +3 -0
  4. test.jsonl +0 -0
  5. train.jsonl +3 -0
  6. validation.jsonl +0 -0
  7. videos_encrypted.sh +1 -0
.gitattributes CHANGED
@@ -58,3 +58,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
 
 
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
61
+ train.jsonl filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,94 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - en
4
+ license: cc-by-nc-4.0
5
+ configs:
6
+ - config_name: default
7
+ data_files:
8
+ - split: validation
9
+ path: validation.jsonl
10
+ - split: train
11
+ path: train.jsonl
12
+ - split: test
13
+ path: test.jsonl
14
+ ---
15
+
16
+ # REFRAMED: Towards Realistic Audio Description Generation for Movies
17
+
18
+ ## Abstract
19
+
20
+ Audio Description (AD) is a verbal narration of key visual content in videos, enabling access for visually impaired audiences. Unlike standard video captioning, AD is a structured editorial task: descriptions must be inserted into gaps in dialogue and must convey only what is needed to understand the narrative being told. However, existing approaches formulate AD generation in an artificial setting where both the content and timing of descriptions are pre-specified, reducing the task to clip-level captioning. They further rely on noisy transcription and alignment pipelines, and lack the rich parallel data required for modeling narrative context. We introduce a new formulation of AD generation in which models must jointly decide what to describe and when to do it. To support this, we present REFRAMED, a high-quality dataset of 2,023 videos that span 3,302 scenes from 206 movies, with professional AD transcripts (both American and British versions), professional subtitles and aligned screenplays. We also provide a manually curated challenge set that pairs full movies with multiple AD references, together with evaluation protocols that leverage dialogue gaps and multi-reference comparisons. Experiments with state-of-the-art AD systems and multimodal LLMs show that they outperform trivial baselines but fall far short of expert human performance. Our dataset and benchmark establish a new foundation for research on video understanding.
21
+
22
+ ## Data format
23
+
24
+ Each row corresponds to one video excerpt. AD and subtitles are provided in SRT format. One option for processing them in Python is [`pysrt`](https://pypi.org/project/pysrt/). Screenplays are provided as JSON-serialized dictionaries, or the string "null" when unavailable. Keys in the dictionaries correspond to scene indices in the [MovieSum](https://huggingface.co/datasets/rohitsaxena/MovieSum) dataset and values contain the aligned scenes’ XML.
25
+
26
+ ## Decryption
27
+
28
+ The test set has been encrypted (following [tell_me_a_story](https://github.com/google-deepmind/tell_me_a_story)). The symmetric key `skey.key` and `private_key.pem` are provided in `keys.zip`. The following example decrypts one value from a loaded dataset (ensure you have [`cryptography`](https://pypi.org/project/cryptography/) installed)
29
+
30
+ ```python
31
+ from datasets import load_dataset
32
+
33
+ from cryptography.hazmat.primitives import hashes
34
+ from cryptography.hazmat.primitives.asymmetric import padding
35
+ from cryptography.hazmat.backends import default_backend
36
+ from cryptography.hazmat.primitives import serialization
37
+ from cryptography.fernet import Fernet
38
+
39
+ import json
40
+
41
+ test = load_dataset("igorsterner/reframed", split="test", streaming=True)
42
+ example = next(iter(test))
43
+
44
+ # Load the private key.
45
+ with open("keys/private_key.pem", "rb") as f:
46
+ private_key = serialization.load_pem_private_key(
47
+ f.read(),
48
+ password=None,
49
+ backend=default_backend()
50
+ )
51
+
52
+ # Load the symmetrical key.
53
+ with open("keys/skey.key", "rb") as f:
54
+ skey = f.read()
55
+
56
+ # Decrypt the symmetrical key.
57
+ unenc_skey = private_key.decrypt(
58
+ skey,
59
+ padding.OAEP(
60
+ mgf=padding.MGF1(algorithm=hashes.SHA256()),
61
+ algorithm=hashes.SHA256(),
62
+ label=None
63
+ )
64
+ )
65
+
66
+ # Decrypt a value as follows.
67
+ f = Fernet(unenc_skey)
68
+ encrypted_value = example["imdb_id"]
69
+ decrypted_value = json.loads(
70
+ f.decrypt(encrypted_value.encode("ascii")).decode("utf-8")
71
+ )
72
+
73
+ print(decrypted_value)
74
+ ```
75
+
76
+ ## Videos
77
+
78
+ You can view a video at `https://www.rottentomatoes.com/m/{rottentomatoes_id}/videos/{video_id}`. See `videos_encrypted.sh` for more. This bash file is encrypted with the same key as above and can be decrypted by applying `f.decrypt(...)` to its contents.
79
+
80
+ ## License
81
+
82
+ Creative Commons Attribution Non Commercial 4.0
83
+
84
+ ## Citation
85
+
86
+ ```bibtex
87
+ @inproceedings{sterner2026reframed,
88
+ title={{REFRAMED}: Towards Realistic Audio Description Generation for Movies},
89
+ author={Igor Sterner and Mirella Lapata and Alex Lascarides and Frank Keller},
90
+ booktitle={Third Conference on Language Modeling},
91
+ year={2026},
92
+ url={https://openreview.net/forum?id=8VyU9qy4Vy}
93
+ }
94
+ ```
keys.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:07e2acf028c0d9544885b3861c4beb6dbd7ff063af1b8416a2652bedb8eb0197
3
+ size 2053
test.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
train.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:60debe0c159d281295a4cb709754ff5839d60a9661a0ff5aea1b618b04efce60
3
+ size 18289365
validation.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
videos_encrypted.sh ADDED
@@ -0,0 +1 @@
 
 
1
+ gAAAAABqtRIw0ZXcsLFKmEgeS9Y7mdXXwUQz9qod2GXcx91HDpH9Bf-sWCRW51F2mDVceth0EnLE0et0ddg4AH0QQe6mUCBC7f3_PC0AqDoYFO9nNZEqb1G7GP6-gqhL1F2bHZ61gLBIslG7gfNYxy3kgU7YTEwEb7uuLXB6ovaJN5LBJP-FE1ooWk_Vqp88wJ3zQc-j5plGp5C_DXrkX4jpWO72wjp0ulycHlBDWOifXzjEYcPk8aVE-_RkGsYN4qrS8yPUU3-IfwdbzOHBWgRaSetybq_5HC8UqSNkA047bXfAHszAxS-RwXBEmDnkU12FGIQMR9bkE-Szsqr0HUUFWKkdn7aAEQ==