| import argparse |
| import json |
| import os |
| from pathlib import Path |
| from VLABench.evaluation.evaluator import VLMEvaluator |
| from VLABench.evaluation.model.vlm import * |
|
|
| def initialize_model(model_name, *args, **kwargs): |
| cls = globals().get(model_name) |
| if cls is None: |
| raise ValueError(f"Model '{model_name}' not found in the current namespace.") |
| |
| return cls(*args, **kwargs) |
|
|
| def parse_args(): |
| parser = argparse.ArgumentParser(description="Run VLM benchmark with specified model and parameters.") |
| parser.add_argument("--vlm_name", type=str, default="GPT_4v", choices=["GPT_4v", "Qwen2_VL", "InternVL2", "MiniCPM_V2_6", "GLM4v", "Llava_NeXT", "Gemini", "Claude"], help="Name of the model class to instantiate") |
| parser.add_argument("--save_interval", type=int, default=1, help="Interval for saving benchmark results") |
| parser.add_argument("--few-shot-num", type=int, default=0, help="Number of few-shot examples") |
| parser.add_argument("--eval-dimension", nargs="+", type=str, default=["M&T", "CommonSense", "Semantic", "Spatial", "PhysicalLaw", "Complex"], help="evaluation dimensions") |
| parser.add_argument("--tasks", nargs='+', default=None, help="Specific tasks to run, default is None, meaning evaluate on all the tasks") |
| parser.add_argument("--n-episodes", type=int, default=100, help="Number of episodes to evaluate for a task") |
| parser.add_argument("--with-cot", default=False, action="store_true", help="Whether to use chain of thought") |
| parser.add_argument("--task-list-json", type=str, default=None, help="Optional JSON file defining the task list") |
| parser.add_argument("--save-dir", type=str, default=None, help="Directory to store evaluation outputs") |
| parser.add_argument("--data-root", type=str, default=None, help="Root directory of the VLM evaluation dataset") |
| return parser.parse_args() |
|
|
| def main(): |
| args = parse_args() |
| assert len(args.eval_dimension) > 0, "Please specify the evaluation dimension" |
| vlabench_root = os.getenv("VLABENCH_ROOT") |
| if not vlabench_root: |
| raise EnvironmentError("VLABENCH_ROOT is not set. Please export it before running the script.") |
|
|
| data_root = Path(args.data_root or os.path.join(vlabench_root, "../dataset")) |
| save_root = Path(args.save_dir or os.path.join(vlabench_root, "../logs/vlm")).resolve() |
| save_root.mkdir(parents=True, exist_ok=True) |
|
|
| for eval_dim in args.eval_dimension: |
| dim_data_path = data_root / f"vlm_evaluation_v1.0/{eval_dim}" |
| if not dim_data_path.exists(): |
| raise FileNotFoundError(f"Data path not found: {dim_data_path}") |
|
|
| if args.tasks is None: |
| task_list = sorted(os.listdir(dim_data_path)) |
| else: |
| task_list = args.tasks |
|
|
| if args.task_list_json is not None: |
| task_list_path = Path(args.task_list_json) |
| if not task_list_path.is_file(): |
| raise FileNotFoundError(f"Task list JSON not found: {task_list_path}") |
| with open(task_list_path, 'r') as f: |
| task_list = json.load(f) |
| if not isinstance(task_list, (list, tuple)): |
| raise ValueError("task_list_json must contain a list of tasks") |
|
|
| evaluator = VLMEvaluator( |
| tasks=task_list, |
| n_episodes=args.n_episodes, |
| data_path=str(dim_data_path), |
| save_path=str(save_root), |
| ) |
| |
| vlm = initialize_model(args.vlm_name) |
|
|
| evaluator.evaluate( |
| vlm, |
| save_interval=args.save_interval, |
| few_shot_num=args.few_shot_num, |
| with_CoT=args.with_cot, |
| eval_dim=eval_dim, |
| ) |
| result = evaluator.get_final_score_dict( |
| args.vlm_name, |
| few_shot_num=args.few_shot_num, |
| with_CoT=args.with_cot, |
| ) |
| model_save_dir = save_root / args.vlm_name |
| model_save_dir.mkdir(parents=True, exist_ok=True) |
| with open(model_save_dir / f"{eval_dim}_result.json", "w") as f: |
| json.dump(result, f, indent=4) |
|
|
| if __name__ == "__main__": |
| main() |
|
|
