Fang Yunhao
commited on
Commit
·
bfd1f01
1
Parent(s):
73121ca
Update.
Browse files- evaluation.py +205 -88
evaluation.py
CHANGED
@@ -1,13 +1,17 @@
|
|
1 |
-
from dataclasses import dataclass
|
2 |
from enum import Enum
|
3 |
from pathlib import Path
|
4 |
from typing import Dict, List, Optional, Union
|
5 |
import logging
|
6 |
import os
|
7 |
-
|
|
|
|
|
|
|
|
|
|
|
8 |
import numpy as np
|
9 |
from mmengine import load, dump
|
10 |
-
from tqdm import tqdm
|
11 |
from collections import defaultdict
|
12 |
|
13 |
|
@@ -17,11 +21,9 @@ class EvaluationType(Enum):
|
|
17 |
COMMON_SENSE = "common_sense"
|
18 |
|
19 |
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
PROMPT_TEMPLATES: Dict[str, str] = {
|
25 |
EvaluationType.INSTRUCTION.value: """
|
26 |
Evaluate if this video follows the instruction: '{instruction}'.
|
27 |
Use the following scoring criteria:
|
@@ -33,17 +35,22 @@ class EvaluationConfig:
|
|
33 |
|
34 |
Let's analyze step-by-step and conclude with 'Score: [score]'.
|
35 |
""".strip(),
|
|
|
36 |
EvaluationType.PHYSICAL_LAWS.value: """
|
37 |
Watch the video and determine if it shows any '{physical_laws}'
|
38 |
Let's think step-by-step and conclude with "Yes" or "No".
|
39 |
""".strip(),
|
|
|
40 |
EvaluationType.COMMON_SENSE.value: """
|
41 |
Does the video exhibit '{common_sense}'?
|
42 |
Let's think step-by-step and conclude with "Yes" or "No".
|
43 |
""".strip(),
|
44 |
}
|
45 |
|
46 |
-
|
|
|
|
|
|
|
47 |
EvaluationType.INSTRUCTION.value: None,
|
48 |
EvaluationType.PHYSICAL_LAWS.value: [
|
49 |
"Violation of Newton's Law: Objects move without any external force.",
|
@@ -59,156 +66,266 @@ class EvaluationConfig:
|
|
59 |
}
|
60 |
|
61 |
|
62 |
-
|
63 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
64 |
|
|
|
|
|
|
|
65 |
def __init__(self, judge_path: str, video_dir: str, config: EvaluationConfig):
|
66 |
self.judge = self._load_judge(judge_path)
|
67 |
self.video_dir = Path(video_dir)
|
68 |
self.config = config
|
69 |
self.logger = logging.getLogger(__name__)
|
|
|
70 |
|
71 |
@staticmethod
|
72 |
def _load_judge(judge_path: str):
|
73 |
-
"""Load the
|
74 |
import llava
|
75 |
-
|
76 |
return llava.load(judge_path)
|
77 |
|
78 |
-
def _load_video(self, video_name: str) -> Optional[
|
79 |
"""Load a video file for evaluation."""
|
80 |
video_path = self.video_dir / f"{video_name}.mp4"
|
81 |
if not video_path.exists():
|
82 |
self.logger.warning(f"Video not found: {video_path}")
|
83 |
return None
|
84 |
import llava
|
85 |
-
|
86 |
return llava.Video(str(video_path))
|
87 |
|
88 |
-
def evaluate_video(
|
89 |
-
self, video: "llava.Video", prompt: str, cot: bool = True
|
90 |
-
) -> str:
|
91 |
"""Generate evaluation content for a video."""
|
92 |
if not cot:
|
93 |
prompt = prompt.replace(
|
94 |
"Let's think step-by-step and conclude with", "Answer with"
|
95 |
-
).replace(
|
|
|
|
|
96 |
return self.judge.generate_content([video, prompt])
|
97 |
|
98 |
def process_results(self, preds: Dict, accs: defaultdict) -> float:
|
99 |
-
"""Process and print evaluation results."""
|
100 |
num_insts = len(preds)
|
101 |
total_score = 0
|
102 |
-
|
103 |
category_mapping = {
|
104 |
2: [("framewise", "temporal")],
|
105 |
-
5: [("newton", "mass", "fluid", "penetration", "gravity")]
|
106 |
}
|
107 |
|
108 |
for category, scores in accs.items():
|
109 |
-
|
110 |
num_sub = len(scores) // num_insts
|
111 |
-
|
112 |
if num_sub == 1:
|
113 |
-
|
114 |
-
|
115 |
-
total_score +=
|
116 |
elif num_sub in category_mapping:
|
117 |
-
sub_scores =
|
118 |
for i, sub in enumerate(category_mapping[num_sub][0]):
|
119 |
sub_mean = np.mean(scores[i::num_sub])
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
125 |
else:
|
126 |
raise ValueError(f"Unexpected number of subcategories: {num_sub}")
|
127 |
|
|
|
128 |
return total_score
|
129 |
|
130 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
131 |
def main():
|
132 |
import argparse
|
133 |
-
|
134 |
parser = argparse.ArgumentParser(description="Evaluate World Model Benchmark")
|
135 |
-
parser.add_argument(
|
136 |
-
|
137 |
-
)
|
138 |
-
parser.add_argument(
|
139 |
-
|
140 |
-
|
141 |
-
parser.add_argument(
|
142 |
-
"--save_name", type=str, required=True, help="Path to save evaluation results"
|
143 |
-
)
|
144 |
-
parser.add_argument(
|
145 |
-
"--cot", action="store_true", help="Enable Chain-of-Thought output"
|
146 |
-
)
|
147 |
-
|
148 |
args = parser.parse_args()
|
149 |
-
|
150 |
-
# Setup logging
|
151 |
-
logging.basicConfig(
|
|
|
|
|
|
|
|
|
152 |
logger = logging.getLogger(__name__)
|
153 |
|
154 |
# Initialize evaluator
|
155 |
config = EvaluationConfig()
|
156 |
evaluator = WorldModelEvaluator(args.judge, args.video_dir, config)
|
157 |
-
|
158 |
-
|
|
|
|
|
159 |
validation_set = load("./worldmodelbench.json")
|
160 |
-
|
161 |
# Check for existing results
|
162 |
-
save_path = f"{args.save_name}_cot" if args.cot else args.save_name
|
163 |
if os.path.exists(save_path):
|
|
|
164 |
results = load(save_path)
|
165 |
try:
|
166 |
preds, accs = results["preds"], results["accs"]
|
167 |
except KeyError:
|
168 |
raise KeyError("Expected keys not found in results file")
|
169 |
else:
|
|
|
170 |
preds = {}
|
171 |
accs = defaultdict(list)
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
|
180 |
-
|
181 |
-
|
182 |
-
|
183 |
-
|
184 |
-
|
185 |
-
|
186 |
-
|
187 |
-
|
188 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
189 |
pred = evaluator.evaluate_video(video, prompt, args.cot)
|
190 |
preds_i.append(pred)
|
191 |
-
|
192 |
-
|
193 |
-
|
194 |
-
|
195 |
-
|
196 |
-
|
197 |
-
|
198 |
-
|
199 |
-
|
200 |
-
|
201 |
-
|
202 |
-
|
203 |
-
|
204 |
-
|
205 |
-
|
206 |
-
|
|
|
|
|
|
|
207 |
|
208 |
# Process and display results
|
|
|
209 |
total_score = evaluator.process_results(preds, accs)
|
210 |
-
print(f"\nTotal score: {total_score:.2f}")
|
211 |
|
212 |
|
213 |
if __name__ == "__main__":
|
214 |
main()
|
|
|
|
1 |
+
from dataclasses import dataclass, field
|
2 |
from enum import Enum
|
3 |
from pathlib import Path
|
4 |
from typing import Dict, List, Optional, Union
|
5 |
import logging
|
6 |
import os
|
7 |
+
from rich.console import Console
|
8 |
+
from rich.table import Table
|
9 |
+
from rich.panel import Panel
|
10 |
+
# from rich.progress import track
|
11 |
+
# from rich import print as rprint
|
12 |
+
from rich.progress import Progress, BarColumn, TimeRemainingColumn
|
13 |
import numpy as np
|
14 |
from mmengine import load, dump
|
|
|
15 |
from collections import defaultdict
|
16 |
|
17 |
|
|
|
21 |
COMMON_SENSE = "common_sense"
|
22 |
|
23 |
|
24 |
+
def get_default_prompt_templates() -> Dict[str, str]:
|
25 |
+
"""Factory function for default prompt templates."""
|
26 |
+
return {
|
|
|
|
|
27 |
EvaluationType.INSTRUCTION.value: """
|
28 |
Evaluate if this video follows the instruction: '{instruction}'.
|
29 |
Use the following scoring criteria:
|
|
|
35 |
|
36 |
Let's analyze step-by-step and conclude with 'Score: [score]'.
|
37 |
""".strip(),
|
38 |
+
|
39 |
EvaluationType.PHYSICAL_LAWS.value: """
|
40 |
Watch the video and determine if it shows any '{physical_laws}'
|
41 |
Let's think step-by-step and conclude with "Yes" or "No".
|
42 |
""".strip(),
|
43 |
+
|
44 |
EvaluationType.COMMON_SENSE.value: """
|
45 |
Does the video exhibit '{common_sense}'?
|
46 |
Let's think step-by-step and conclude with "Yes" or "No".
|
47 |
""".strip(),
|
48 |
}
|
49 |
|
50 |
+
|
51 |
+
def get_default_question_pool() -> Dict[str, Optional[List[str]]]:
|
52 |
+
"""Factory function for default question pool."""
|
53 |
+
return {
|
54 |
EvaluationType.INSTRUCTION.value: None,
|
55 |
EvaluationType.PHYSICAL_LAWS.value: [
|
56 |
"Violation of Newton's Law: Objects move without any external force.",
|
|
|
66 |
}
|
67 |
|
68 |
|
69 |
+
@dataclass
|
70 |
+
class EvaluationConfig:
|
71 |
+
"""Configuration for evaluation prompts and scoring criteria."""
|
72 |
+
PROMPT_TEMPLATES: Dict[str, str] = field(default_factory=get_default_prompt_templates)
|
73 |
+
QUESTION_POOL: Dict[str, Optional[List[str]]] = field(default_factory=get_default_question_pool)
|
74 |
+
|
75 |
+
|
76 |
+
class ResultsPrinter:
|
77 |
+
"""Handles formatted output of evaluation results."""
|
78 |
+
|
79 |
+
def __init__(self):
|
80 |
+
self.console = Console()
|
81 |
+
|
82 |
+
def print_header(self, text: str):
|
83 |
+
"""Print a styled header."""
|
84 |
+
self.console.print(f"\n[bold blue]{text}[/bold blue]")
|
85 |
+
|
86 |
+
def print_score(self, category: str, score: float, indent: int = 0):
|
87 |
+
"""Print a score with proper formatting."""
|
88 |
+
indent_str = " " * indent
|
89 |
+
self.console.print(f"{indent_str}[cyan]{category}:[/cyan] [yellow]{score:.2f}[/yellow]")
|
90 |
+
|
91 |
+
def create_results_table(self, category: str, scores: Dict[str, float]) -> Table:
|
92 |
+
"""Create a rich table for displaying results."""
|
93 |
+
table = Table(title=f"{category} Results", show_header=True, header_style="bold magenta")
|
94 |
+
table.add_column("Metric", style="cyan")
|
95 |
+
table.add_column("Score", justify="right", style="yellow")
|
96 |
+
|
97 |
+
for metric, score in scores.items():
|
98 |
+
table.add_row(metric, f"{score:.2f}")
|
99 |
+
|
100 |
+
return table
|
101 |
+
|
102 |
+
def print_summary_panel(self, total_score: float, num_categories: int):
|
103 |
+
"""Print a panel with summary information."""
|
104 |
+
panel = Panel(
|
105 |
+
f"[bold green]Total Score: {total_score:.2f}[/bold green]\n",
|
106 |
+
# f"[blue]Average per category: {total_score/num_categories:.2f}[/blue]",
|
107 |
+
title="Evaluation Summary",
|
108 |
+
border_style="green"
|
109 |
+
)
|
110 |
+
self.console.print(panel)
|
111 |
+
|
112 |
|
113 |
+
class WorldModelEvaluator:
|
114 |
+
"""Evaluates world model benchmark videos using VILA model."""
|
115 |
+
|
116 |
def __init__(self, judge_path: str, video_dir: str, config: EvaluationConfig):
|
117 |
self.judge = self._load_judge(judge_path)
|
118 |
self.video_dir = Path(video_dir)
|
119 |
self.config = config
|
120 |
self.logger = logging.getLogger(__name__)
|
121 |
+
self.printer = ResultsPrinter()
|
122 |
|
123 |
@staticmethod
|
124 |
def _load_judge(judge_path: str):
|
125 |
+
"""Load the VILA judge model."""
|
126 |
import llava
|
|
|
127 |
return llava.load(judge_path)
|
128 |
|
129 |
+
def _load_video(self, video_name: str) -> Optional['llava.Video']:
|
130 |
"""Load a video file for evaluation."""
|
131 |
video_path = self.video_dir / f"{video_name}.mp4"
|
132 |
if not video_path.exists():
|
133 |
self.logger.warning(f"Video not found: {video_path}")
|
134 |
return None
|
135 |
import llava
|
|
|
136 |
return llava.Video(str(video_path))
|
137 |
|
138 |
+
def evaluate_video(self, video: 'llava.Video', prompt: str, cot: bool = True) -> str:
|
|
|
|
|
139 |
"""Generate evaluation content for a video."""
|
140 |
if not cot:
|
141 |
prompt = prompt.replace(
|
142 |
"Let's think step-by-step and conclude with", "Answer with"
|
143 |
+
).replace(
|
144 |
+
"Let's analyze step-by-step and conclude with", "Answer with"
|
145 |
+
)
|
146 |
return self.judge.generate_content([video, prompt])
|
147 |
|
148 |
def process_results(self, preds: Dict, accs: defaultdict) -> float:
|
149 |
+
"""Process and print evaluation results with rich formatting."""
|
150 |
num_insts = len(preds)
|
151 |
total_score = 0
|
152 |
+
|
153 |
category_mapping = {
|
154 |
2: [("framewise", "temporal")],
|
155 |
+
5: [("newton", "mass", "fluid", "penetration", "gravity")]
|
156 |
}
|
157 |
|
158 |
for category, scores in accs.items():
|
159 |
+
self.printer.print_header(f"{category.replace('_', ' ').title()} Details")
|
160 |
num_sub = len(scores) // num_insts
|
161 |
+
|
162 |
if num_sub == 1:
|
163 |
+
overall = np.mean(scores)
|
164 |
+
self.printer.print_score("Overall", overall)
|
165 |
+
total_score += overall
|
166 |
elif num_sub in category_mapping:
|
167 |
+
sub_scores = {}
|
168 |
for i, sub in enumerate(category_mapping[num_sub][0]):
|
169 |
sub_mean = np.mean(scores[i::num_sub])
|
170 |
+
sub_scores[sub.title()] = sub_mean
|
171 |
+
|
172 |
+
# Create and display results table
|
173 |
+
table = self.printer.create_results_table(
|
174 |
+
category.replace('_', ' ').title(),
|
175 |
+
sub_scores
|
176 |
+
)
|
177 |
+
self.printer.console.print(table)
|
178 |
+
|
179 |
+
overall = np.sum(list(sub_scores.values()))
|
180 |
+
self.printer.print_score("Overall", overall, indent=2)
|
181 |
+
total_score += overall
|
182 |
else:
|
183 |
raise ValueError(f"Unexpected number of subcategories: {num_sub}")
|
184 |
|
185 |
+
self.printer.print_summary_panel(total_score, len(accs))
|
186 |
return total_score
|
187 |
|
188 |
|
189 |
+
def save_results(results: Dict, save_path: str):
|
190 |
+
"""Save evaluation results to a file."""
|
191 |
+
dump(results, save_path, indent=4)
|
192 |
+
Console().print(f"[green]Results saved to: {save_path}[/green]")
|
193 |
+
|
194 |
+
class RichLogHandler(logging.Handler):
|
195 |
+
"""Custom logging handler that uses Rich for formatting."""
|
196 |
+
def __init__(self):
|
197 |
+
super().__init__()
|
198 |
+
self.console = Console()
|
199 |
+
|
200 |
+
def emit(self, record):
|
201 |
+
try:
|
202 |
+
msg = self.format(record)
|
203 |
+
style = "bold red" if record.levelno >= logging.WARNING else "blue"
|
204 |
+
self.console.print(f"[{style}]{msg}[/{style}]")
|
205 |
+
except Exception:
|
206 |
+
self.handleError(record)
|
207 |
+
|
208 |
def main():
|
209 |
import argparse
|
210 |
+
|
211 |
parser = argparse.ArgumentParser(description="Evaluate World Model Benchmark")
|
212 |
+
parser.add_argument("--judge", type=str, required=True, help="Path to judge model checkpoint")
|
213 |
+
parser.add_argument("--video_dir", type=str, required=True, help="Path to generated video directory")
|
214 |
+
parser.add_argument("--save_name", type=str, required=True, help="Path to save evaluation results")
|
215 |
+
parser.add_argument("--cot", action="store_true", help="Enable Chain-of-Thought output")
|
216 |
+
parser.add_argument("--no-save", action="store_true", help="Disable saving results")
|
217 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
218 |
args = parser.parse_args()
|
219 |
+
|
220 |
+
# Setup logging with custom Rich handler
|
221 |
+
logging.basicConfig(
|
222 |
+
level=logging.INFO,
|
223 |
+
format="%(message)s",
|
224 |
+
handlers=[RichLogHandler()]
|
225 |
+
)
|
226 |
logger = logging.getLogger(__name__)
|
227 |
|
228 |
# Initialize evaluator
|
229 |
config = EvaluationConfig()
|
230 |
evaluator = WorldModelEvaluator(args.judge, args.video_dir, config)
|
231 |
+
printer = ResultsPrinter()
|
232 |
+
|
233 |
+
# Load validation set with status message
|
234 |
+
printer.console.print("[bold]Loading validation set...[/bold]")
|
235 |
validation_set = load("./worldmodelbench.json")
|
236 |
+
|
237 |
# Check for existing results
|
238 |
+
save_path = f"{args.save_name}_cot.json" if args.cot else f"{args.save_name}.json"
|
239 |
if os.path.exists(save_path):
|
240 |
+
printer.console.print("[bold yellow]Loading existing results...[/bold yellow]")
|
241 |
results = load(save_path)
|
242 |
try:
|
243 |
preds, accs = results["preds"], results["accs"]
|
244 |
except KeyError:
|
245 |
raise KeyError("Expected keys not found in results file")
|
246 |
else:
|
247 |
+
printer.console.print("[bold green]Starting new evaluation...[/bold green]")
|
248 |
preds = {}
|
249 |
accs = defaultdict(list)
|
250 |
+
|
251 |
+
# Create a single progress instance for all operations
|
252 |
+
with Progress(
|
253 |
+
"[progress.description]{task.description}",
|
254 |
+
BarColumn(),
|
255 |
+
"[progress.percentage]{task.percentage:>3.0f}%",
|
256 |
+
TimeRemainingColumn(),
|
257 |
+
console=printer.console
|
258 |
+
) as progress:
|
259 |
+
# Main task for video processing
|
260 |
+
video_task = progress.add_task("Processing videos", total=len(validation_set))
|
261 |
+
|
262 |
+
for vid, v_i in enumerate(validation_set):
|
263 |
+
video_name = Path(v_i["first_frame"]).stem
|
264 |
+
video = evaluator._load_video(video_name)
|
265 |
+
if not video:
|
266 |
+
progress.advance(video_task)
|
267 |
+
continue
|
268 |
+
|
269 |
+
# Evaluation task
|
270 |
+
eval_task = progress.add_task(
|
271 |
+
f"Evaluating {video_name}",
|
272 |
+
total=len(EvaluationType)
|
273 |
+
)
|
274 |
+
|
275 |
+
for eval_type in EvaluationType:
|
276 |
+
preds_i = []
|
277 |
+
prompt_template = config.PROMPT_TEMPLATES[eval_type.value]
|
278 |
+
questions = config.QUESTION_POOL[eval_type.value]
|
279 |
+
|
280 |
+
if questions:
|
281 |
+
accs_i = []
|
282 |
+
# Questions task
|
283 |
+
question_task = progress.add_task(
|
284 |
+
f"Processing {eval_type.value} questions",
|
285 |
+
total=len(questions)
|
286 |
+
)
|
287 |
+
|
288 |
+
for question in questions:
|
289 |
+
format_kwargs = {
|
290 |
+
f"{eval_type.value}": question.lower()
|
291 |
+
}
|
292 |
+
prompt = prompt_template.format(**format_kwargs)
|
293 |
+
pred = evaluator.evaluate_video(video, prompt, args.cot)
|
294 |
+
preds_i.append(pred)
|
295 |
+
accs_i.append("no" in pred.lower())
|
296 |
+
progress.advance(question_task)
|
297 |
+
|
298 |
+
progress.remove_task(question_task)
|
299 |
+
accs[eval_type.value].extend(accs_i)
|
300 |
+
else:
|
301 |
+
prompt = prompt_template.format(instruction=v_i["text_instruction"])
|
302 |
pred = evaluator.evaluate_video(video, prompt, args.cot)
|
303 |
preds_i.append(pred)
|
304 |
+
try:
|
305 |
+
score = float(pred.split(":")[-1].strip(" ."))
|
306 |
+
except ValueError:
|
307 |
+
logger.warning(f"Could not parse score from prediction: {pred}")
|
308 |
+
score = 0
|
309 |
+
accs[eval_type.value].append(score)
|
310 |
+
|
311 |
+
if video_name not in preds:
|
312 |
+
preds[video_name] = {}
|
313 |
+
preds[video_name][eval_type.value] = preds_i
|
314 |
+
progress.advance(eval_task)
|
315 |
+
|
316 |
+
progress.remove_task(eval_task)
|
317 |
+
progress.advance(video_task)
|
318 |
+
|
319 |
+
# Save results if requested
|
320 |
+
if not args.no_save:
|
321 |
+
results = {"preds": preds, "accs": accs}
|
322 |
+
save_results(results, save_path)
|
323 |
|
324 |
# Process and display results
|
325 |
+
printer.console.print("\n[bold]Final Evaluation Results[/bold]")
|
326 |
total_score = evaluator.process_results(preds, accs)
|
|
|
327 |
|
328 |
|
329 |
if __name__ == "__main__":
|
330 |
main()
|
331 |
+
|