Ziyang Wang commited on
Commit
6375fde
·
1 Parent(s): 1bf5b23

add paper-baseline seed script

Browse files
Files changed (1) hide show
  1. scripts/seed_paper_baselines.py +130 -0
scripts/seed_paper_baselines.py ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Seed the public leaderboard dataset with the paper's Table 1 baselines.
2
+
3
+ Generates one JSON record per system under seeds/submissions/<uuid>.json.
4
+ UUIDs are deterministic (uuid5 of the method name), so re-running this script
5
+ overwrites existing seed records rather than creating duplicates.
6
+
7
+ To publish to the public dataset, either:
8
+ (a) Drag-drop the resulting folder onto the HF UI of
9
+ Ted412/EgoMemReason-Leaderboard → submissions/, or
10
+ (b) Run:
11
+ from huggingface_hub import HfApi
12
+ HfApi().upload_folder(
13
+ folder_path="seeds/submissions",
14
+ path_in_repo="submissions",
15
+ repo_id="Ted412/EgoMemReason-Leaderboard",
16
+ repo_type="dataset",
17
+ )
18
+ """
19
+
20
+ import json
21
+ import pathlib
22
+ import uuid
23
+
24
+ # Stable seed namespace so uuid5 ids are reproducible across runs.
25
+ SEED_NS = uuid.UUID("5a4f2b1c-0000-4ab2-9c01-deadbeef0001")
26
+
27
+ # Authors uploaded the paper's numbers under this HF handle.
28
+ SEED_HF_USER = "Ted412"
29
+ SEED_DATE = "2026-05-01T00:00:00Z" # COLM submission date — adjust if needed.
30
+
31
+ # Source of truth: Table 1 of the paper PDF (Long_Video_Memory_for_Arxiv_v24.pdf).
32
+ # Columns: Tracking, Counting, Ordering, Linking, Spatial, Activity, Overall.
33
+ PAPER_TABLE = [
34
+ # (method_name, category, model_size, uses_external_data, modality,
35
+ # tracking, counting, ordering, linking, spatial, activity, overall)
36
+ ("Random", "Baseline", "—", False, "other",
37
+ 19.6, 16.7, 11.1, 17.3, 19.3, 19.2, 16.8),
38
+
39
+ ("InternVL3.5-8B", "General MLLM", "8B", False, "frames-only",
40
+ 23.0, 29.0, 23.0, 27.0, 34.0, 42.0, 28.0),
41
+ ("Qwen-3-VL-8B", "General MLLM", "8B", False, "frames-only",
42
+ 35.0, 28.0, 23.0, 21.0, 40.0, 42.0, 29.6),
43
+ ("InternVL3.5-38B", "General MLLM", "38B", False, "frames-only",
44
+ 33.0, 40.0, 27.0, 24.0, 46.0, 32.0, 32.6),
45
+ ("Qwen-3-VL-30B-A3B", "General MLLM", "30B MoE", False, "frames-only",
46
+ 36.0, 48.0, 25.0, 26.0, 40.0, 30.0, 34.0),
47
+ ("Qwen-3-VL-32B", "General MLLM", "32B", False, "frames-only",
48
+ 35.0, 46.0, 27.0, 27.0, 50.0, 46.0, 36.8),
49
+ ("GPT-5", "General MLLM", "API", False, "frames-only",
50
+ 29.0, 42.0, 20.0, 18.0, 32.0, 28.0, 27.8),
51
+ ("Gemini-3-Flash", "General MLLM", "API", False, "frames-only",
52
+ 46.0, 28.0, 36.0, 44.0, 44.0, 44.0, 39.6),
53
+ ("Gemini-3.1-Pro", "General MLLM", "API", False, "frames-only",
54
+ 40.0, 26.0, 44.0, 33.0, 40.0, 48.0, 37.4),
55
+
56
+ ("LongVA-7B", "Video-specific MLLM", "7B", False, "frames-only",
57
+ 22.0, 18.0, 20.0, 20.0, 20.0, 22.0, 20.6),
58
+ ("StreamingVLM", "Video-specific MLLM", "—", False, "video-only",
59
+ 25.0, 29.0, 21.0, 20.0, 20.0, 32.0, 24.2),
60
+ ("InternVideo-2.5-8B", "Video-specific MLLM", "8B", False, "frames-only",
61
+ 29.0, 27.0, 25.0, 15.0, 32.0, 32.0, 25.6),
62
+ ("VideoLLaMA3-8B", "Video-specific MLLM", "8B", False, "frames-only",
63
+ 23.0, 31.0, 27.0, 32.0, 38.0, 36.0, 30.0),
64
+ ("Molmo2-8B", "Video-specific MLLM", "8B", False, "frames-only",
65
+ 36.0, 50.0, 27.0, 25.0, 34.0, 22.0, 33.2),
66
+
67
+ ("SiLVR", "Agentic Video Framework", "API + 8B", False, "captions-only",
68
+ 31.0, 14.0, 27.0, 17.0, 18.0, 28.0, 22.4),
69
+ ("Ego-R1", "Agentic Video Framework", "API", False, "frames-only",
70
+ 30.0, 18.0, 23.0, 18.0, 48.0, 32.0, 25.8),
71
+ ("WorldMM", "Agentic Video Framework", "API", False, "frames-only",
72
+ 32.0, 44.0, 21.0, 21.0, 34.0, 36.0, 30.6),
73
+ ("AVP", "Agentic Video Framework", "API", False, "frames-only",
74
+ 34.0, 42.0, 31.0, 27.0, 38.0, 34.0, 34.0),
75
+ ]
76
+
77
+ PROJECT_URL = "https://github.com/Ted412/EgoMemReason"
78
+ PUBLICATION_URL = "https://arxiv.org/abs/<arxiv_id>" # placeholder — fill in after posting
79
+
80
+
81
+ def build_record(row):
82
+ (name, category, size, uses_ext, modality,
83
+ tracking, counting, ordering, linking, spatial, activity, overall) = row
84
+ sid = str(uuid.uuid5(SEED_NS, f"paper-baseline:{name}"))
85
+ return {
86
+ "submission_id": sid,
87
+ "submitted_at_utc": SEED_DATE,
88
+ "hf_user_id": SEED_HF_USER,
89
+ "team_name": f"EgoMemReason paper · {category}",
90
+ "method_name": name,
91
+ "model_size": size,
92
+ "uses_external_data": uses_ext,
93
+ "uses_video_frames": modality,
94
+ "method_description": f"Baseline from Table 1 of the EgoMemReason paper ({category}).",
95
+ "project_url": PROJECT_URL,
96
+ "publication_url": PUBLICATION_URL,
97
+ "is_selected": True,
98
+ "metrics": {
99
+ "Cumulative State Tracking": tracking,
100
+ "Temporal Counting": counting,
101
+ "Event Ordering": ordering,
102
+ "Event Linking": linking,
103
+ "Spatial Preference": spatial,
104
+ "Activity Pattern": activity,
105
+ "Overall": overall,
106
+ },
107
+ }
108
+
109
+
110
+ def main():
111
+ out_dir = pathlib.Path(__file__).resolve().parents[1] / "seeds" / "submissions"
112
+ out_dir.mkdir(parents=True, exist_ok=True)
113
+
114
+ for row in PAPER_TABLE:
115
+ record = build_record(row)
116
+ path = out_dir / f"{record['submission_id']}.json"
117
+ path.write_text(json.dumps(record, indent=2))
118
+
119
+ print(f"Wrote {len(PAPER_TABLE)} seed records to {out_dir}/")
120
+ print("Next: upload to Ted412/EgoMemReason-Leaderboard via HF UI or:")
121
+ print(" python -c \"from huggingface_hub import HfApi; "
122
+ "HfApi().upload_folder("
123
+ "folder_path='seeds/submissions', "
124
+ "path_in_repo='submissions', "
125
+ "repo_id='Ted412/EgoMemReason-Leaderboard', "
126
+ "repo_type='dataset')\"")
127
+
128
+
129
+ if __name__ == "__main__":
130
+ main()