JinPLU commited on
Commit
93f56dd
·
verified ·
1 Parent(s): 7177c6f

Sync ecosystem table and benchmark CLI / research-release cards

Browse files
Files changed (1) hide show
  1. README.md +48 -15
README.md CHANGED
@@ -38,10 +38,13 @@ Naive consistency metrics reward models that ignore cuts and **copy the same bac
38
  |---|---|---|
39
  | [`JinPLu/CaT-Movie`](https://github.com/JinPLu/CaT-Movie) (GitHub) | Core | Training, inference, data preparation, planner & CaT-MBench code |
40
  | [`JinPLU/CaT-Movie`](https://huggingface.co/JinPLU/CaT-Movie) | Core | Renderer checkpoint (`step-3000.safetensors`) |
41
- | [`JinPLU/CaT-Movie-Data`](https://huggingface.co/JinPLU/CaT-Movie-Data) | Core | 75,461 annotated story blocks & 96,000 stratified training samples |
42
- | [`JinPLU/CaT-MBench`](https://huggingface.co/JinPLU/CaT-MBench) *(this repo)* | Core | 122 stories, cut-gated multi-shot VQA bank, verdicts & human labels |
43
- | [`JinPLU/CaT-MBench-Videos`](https://huggingface.co/JinPLU/CaT-MBench-Videos) | Supporting | 736 generated videos (`92 test stories × 8 methods`) |
44
- | [`JinPLU/CaT-MBench-Author-Labels`](https://huggingface.co/JinPLU/CaT-MBench-Author-Labels) | Archive | Nov 6 single-author calibration archive *(superseded by `human/` here)* |
 
 
 
45
 
46
  ---
47
 
@@ -97,7 +100,8 @@ CaT-MBench/
97
  │ └── readout_v17.json # Exact paper Table 1 aggregate metrics & row definitions
98
  ├── judge/
99
  │ ├── cut/{pairs,verdicts}/ # Cut-execution pair inputs & VLM verdicts for all 8 methods
100
- │ └── ... # Whole-video VQA answers for all 8 methods
 
101
  └── human/
102
  ├── human_labels.json # Final dual-annotator blind human judgements (791 total; 557 reviewed)
103
  └── agreement.json # Inter-annotator agreement & human-vs-judge calibration
@@ -116,32 +120,61 @@ hf download JinPLU/CaT-MBench --repo-type dataset --local-dir data/cat_mbench
116
  hf download JinPLU/CaT-MBench-Videos --repo-type dataset --local-dir data/cat_mbench_videos
117
  ```
118
 
119
- ### 2. Option A — Recompute paper tables from shipped judge verdicts (No GPU / API cost)
120
 
121
  From [`JinPLu/CaT-Movie`](https://github.com/JinPLu/CaT-Movie):
122
 
123
  ```bash
124
- PYTHONPATH=src python -m mspc.evaluation.benchmark summarize \
125
- --manifest path/to/eval_manifest.json \
126
- --output outputs/eval/cat_mbench_summary
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
127
  ```
128
 
129
  ### 3. Option B — Score your own model's generated videos
130
 
131
- Generate one `.mp4` per story ID in `stories.json` (`test` split), add your method entry to an evaluation manifest, and run:
132
 
133
  ```bash
134
- PYTHONPATH=src python -m mspc.evaluation.benchmark score \
135
- --manifest path/to/custom_manifest.json \
136
- --output outputs/eval/custom_method \
137
- --components all
 
 
 
 
 
 
 
 
 
 
 
 
 
138
  ```
139
 
140
  ---
141
 
142
  ## Licensing
143
 
144
- Released under **[CC BY-NC-SA 4.0](LICENSE)** (annotations, stories, questions, judge verdicts, human labels, and readouts). Source reference film stills are not redistributed.
145
 
146
  ## Citation
147
 
 
38
  |---|---|---|
39
  | [`JinPLu/CaT-Movie`](https://github.com/JinPLu/CaT-Movie) (GitHub) | Core | Training, inference, data preparation, planner & CaT-MBench code |
40
  | [`JinPLU/CaT-Movie`](https://huggingface.co/JinPLU/CaT-Movie) | Core | Renderer checkpoint (`step-3000.safetensors`) |
41
+ | [`JinPLU/CaT-Movie-Data`](https://huggingface.co/datasets/JinPLU/CaT-Movie-Data) | Core | 75,461 annotated story blocks, `raw_annotations/story_annotation_v2`, member manifests & `96,000` training samples |
42
+ | [`JinPLU/CaT-Movie-T5-Cache`](https://huggingface.co/datasets/JinPLU/CaT-Movie-T5-Cache) | Core | Precomputed `umT5-XXL` BF16 text caches (`332,516` unique prompts) |
43
+ | [`JinPLU/CaT-Movie-VAE-Caches`](https://huggingface.co/datasets/JinPLU/CaT-Movie-VAE-Caches) | Core | Precomputed target VAE latents (`Wan2.2 704p` + `Wan2.1 480p`) |
44
+ | [`JinPLU/CaT-Movie-Source-Videos`](https://huggingface.co/datasets/JinPLU/CaT-Movie-Source-Videos) | Supporting | Complete `18,173` referenced source MP4 clips |
45
+ | [`JinPLU/CaT-MBench`](https://huggingface.co/datasets/JinPLU/CaT-MBench) *(this repo)* | Core | 122 stories, cut-gated multi-shot VQA bank, verdicts & human labels |
46
+ | [`JinPLU/CaT-MBench-Videos`](https://huggingface.co/datasets/JinPLU/CaT-MBench-Videos) | Supporting | 736 generated videos (`92 test stories × 8 methods`) |
47
+ | [`JinPLU/CaT-MBench-Author-Labels`](https://huggingface.co/datasets/JinPLU/CaT-MBench-Author-Labels) | Archive | Nov 6 single-author calibration archive *(superseded by `human/` here)* |
48
 
49
  ---
50
 
 
100
  │ └── readout_v17.json # Exact paper Table 1 aggregate metrics & row definitions
101
  ├── judge/
102
  │ ├── cut/{pairs,verdicts}/ # Cut-execution pair inputs & VLM verdicts for all 8 methods
103
+ │ ├── pairs/{pairs,receipts}/ # 6-frame shot-pair consistency inputs & judge receipts
104
+ │ └── video/rows/ # Whole-video 2 fps event/state answers & prepared metadata
105
  └── human/
106
  ├── human_labels.json # Final dual-annotator blind human judgements (791 total; 557 reviewed)
107
  └── agreement.json # Inter-annotator agreement & human-vs-judge calibration
 
120
  hf download JinPLU/CaT-MBench-Videos --repo-type dataset --local-dir data/cat_mbench_videos
121
  ```
122
 
123
+ ### 2. Option A — Recompute Table 1 & Paired Bootstrap CIs (CPU, No GPU / API Cost)
124
 
125
  From [`JinPLu/CaT-Movie`](https://github.com/JinPLu/CaT-Movie):
126
 
127
  ```bash
128
+ # Recompute exact Table 1 scores from shipped cut & VQA judge verdicts
129
+ PYTHONPATH=src python -m mspc.evaluation.checklist_scoring readout-v17 \
130
+ --bank data/cat_mbench/questions.v17.json \
131
+ --cut-items data/cat_mbench/cut_items.v17.json \
132
+ --cut-root data/cat_mbench/judge/cut \
133
+ --pair-root data/cat_mbench/judge/pairs \
134
+ --video-root data/cat_mbench/judge/video \
135
+ --rows wan22-final-test92-planner-27b-seed42 test92-holocine test92-multishotmaster \
136
+ test92-ic-lora test92-storydiffusion test92-storymem \
137
+ test92-official-wan test92-wan21-1p3b-native \
138
+ --answer-scope existing \
139
+ --output outputs/eval/readout_v17_recomputed.json
140
+
141
+ # Film-clustered paired bootstrap intervals (CaT-Movie vs HoloCine example)
142
+ PYTHONPATH=src python -m mspc.evaluation.checklist_scoring paired \
143
+ --readout outputs/eval/readout_v17_recomputed.json \
144
+ --requests configs/benchmark/story_requests_v4.json \
145
+ --pair test92-holocine wan22-final-test92-planner-27b-seed42 \
146
+ --output outputs/eval/paired_vs_holocine.json
147
  ```
148
 
149
  ### 3. Option B — Score your own model's generated videos
150
 
151
+ Generate one `.mp4` per story ID in `stories.json` (`test` split), run TransNetV2 shot-boundary detection (`checklist_scoring detect-cuts`), and execute the two-stage Google Batch VLM evaluation pipeline (`cut_execution` -> `consistency_pair_batch` + `gemini_checklist_batch`) before calling `checklist_scoring readout-v17`:
152
 
153
  ```bash
154
+ # Stage 1: Cut execution gate (TransNetV2 + mid-frame correlation + 4-frame VLM check)
155
+ PYTHONPATH=src python -m mspc.evaluation.cut_execution prepare ...
156
+ PYTHONPATH=src python -m mspc.evaluation.cut_execution submit ...
157
+ PYTHONPATH=src python -m mspc.evaluation.cut_execution collect ...
158
+
159
+ # Stage 2a: Shot-pair consistency (6 native-resolution frames for character/object/scene)
160
+ PYTHONPATH=src python -m mspc.evaluation.consistency_pair_batch prepare ...
161
+ PYTHONPATH=src python -m mspc.evaluation.consistency_pair_batch submit ...
162
+ PYTHONPATH=src python -m mspc.evaluation.consistency_pair_batch collect ...
163
+
164
+ # Stage 2b: Whole-video events & states (2 fps full video)
165
+ PYTHONPATH=src python -m mspc.evaluation.gemini_checklist_batch prepare ...
166
+ PYTHONPATH=src python -m mspc.evaluation.gemini_checklist_batch submit ...
167
+ PYTHONPATH=src python -m mspc.evaluation.gemini_checklist_batch collect ...
168
+
169
+ # Stage 3: Aggregate cut-gated scores under bank v17
170
+ PYTHONPATH=src python -m mspc.evaluation.checklist_scoring readout-v17 --answer-scope complete ...
171
  ```
172
 
173
  ---
174
 
175
  ## Licensing
176
 
177
+ Released under **[CC BY-NC-SA 4.0](LICENSE)** strictly for non-commercial academic research (annotations, stories, questions, judge verdicts, human labels, and readouts). Referenced source video clips are released separately in [`JinPLU/CaT-Movie-Source-Videos`](https://huggingface.co/datasets/JinPLU/CaT-Movie-Source-Videos) for academic research.
178
 
179
  ## Citation
180