ESTP-Bench/estp_dataset/benchmark/eval.py

import json,os
import numpy as np

def load_multiple_json(file_path):
    """读取包含多个 JSON 对象的文件，并将每个 JSON 对象解析成 Python 对象，存放在列表中。"""
    with open(file_path, 'r', encoding='utf-8') as f:
        content = f.read()
    
    decoder = json.JSONDecoder()
    pos = 0
    results = []
    content_length = len(content)
    
    while pos < content_length:
        # 跳过空白字符
        while pos < content_length and content[pos].isspace():
            pos += 1
        if pos >= content_length:
            break
        try:
            obj, new_pos = decoder.raw_decode(content, pos)
            results.append(obj)
            pos = new_pos
        except json.JSONDecodeError as e:
            # 出现解析错误则退出循环
            print(f"JSON 解析错误: {e}")
            break
    return results


import argparse

def parse_args():
    parser = argparse.ArgumentParser()
    parser.add_argument('--eval_file', type=str, default='')
    parser.add_argument('--eval_mode', type=str, default='all')
    return parser.parse_args()

args = parse_args()

# eval_file = '/root/videollm-online/data/estp_dataset/result/estp_bench_sq_MiniCPMV_passiveevaluator_llama_5_5.json' # tmp_predict_VideollmOnline_v2_correctness
eval_file = '/root/videollm-online/data/estp_dataset/result/estp_bench_sq_MiniCPMV_passiveevaluator_deepseek_1_2.json'
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline_5cases/MiniCPMV_fbf_5casesevaluator_deepseek_5_5.json'
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline_5cases/MiniCPMV_fbf_5cases_0.175evaluator_deepseek_5_5.json'
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline_5cases/MiniCPMV_fbf_5cases_0.175evaluator_deepseek_1_2.json'
eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/MiniCPMV_fbf_singleQA_0.175_v2evaluator_deepseek_1_2.json'
eval_model = 'MiniCPMV'

# eval_file = '/root/videollm-online/data/estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_1_2.json'
# eval_file = '/root/videollm-online/data/estp_dataset/estp_bench_sq_VideollmOnline0.8evaluator_deepseek_1_2.json'
# # eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours/LIVE_IT0.95.json'
# eval_model = 'VideollmOnline' # VideollmOnline MiniCPMV

# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/MMDuetevaluator_deepseek_1_2.json'
# eval_model = 'MMDuet'

# eval_file = '/root/videollm-online/data/estp_dataset/result/estp_bench_sq_EWO_frame_by_frame_v6_fusion_dinov2evaluator_llama_5_5.json'
# eval_file = '/root/videollm-online/data/estp_dataset/result/estp_bench_sq_EWO_frame_by_frame_fitVal_5_cases_v2evaluator_llama_5_5.json'
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOFitValStage3Firstevaluator_llama_5_5.json'
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOInDomainITstage2evaluator_llama_5_5.json'
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq_fbf_EWOInDomainITstage2evaluator_llama_5_5.json'

# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOFitValStage3First0.6evaluator_llama_5_5.json'
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOFitValStage3First0.7evaluator_llama_5_5.json'
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOFitValStage3highv2evaluator_llama_5_5.json'
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOFitValStage3firstv2evaluator_llama_5_5.json'
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOFitValStage3HighRegion_evaluator_llama_5_5.json'
# eval_file = '/root/videollm-online/data/estp_dataset/train_/estpBenchSq5Cases_fbf_beaconlivel_h_stage2_v2evaluator_deepseek_5_5.json'
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours_5cases/LivebaseStage3_high11_evaluator_deepseek_1_2.json'
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours_5cases/LivebaseStage2evaluator_deepseek_5_5.json'
eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours_5cases/LivebaseStage3_high0.31_11evaluator_deepseek_1_2.json'
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours_5cases/LivebaseStage3_high11_evaluator_deepseek_1_2.json'

# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours/LivebaseStage3_high11_evaluator_deepseek_1_2.json'
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours/LivebaseStage2evaluator_deepseek_1_2.json'
eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours/LivebaseStage3_high0.31_1_lowevaluator_deepseek_1_2.json'
eval_model = 'EWO'

# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/LLaVAOneVision_passiveevaluator_deepseek_1_2.json'
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/LLaVAOneVision_fbf_singleQA_0.175evaluator_deepseek_1_2.json'

# eval_model = 'LLaVAOneVision'

# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/LLaVANextVideo7B_passiveevaluator_deepseek_1_2.json'
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/LLaVANextVideo7B_fbf_singleQA_0.175evaluator_deepseek_1_2.json'
# eval_model = 'LLaVANextVideo7B'

# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline_5cases/Qwen2VL_fbf_5casesevaluator_deepseek_5_5.json'
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/Qwen2VL_fbf_singleQA_0.175evaluator_deepseek_1_2.json'
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/Qwen2VL_passiveevaluator_deepseek_1_2.json'
# eval_model = 'Qwen2VL'

# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json'
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/InternVLV28_passiveevaluator_deepseek_1_2.json'
# eval_model = 'InternVLV28'


# eval_model = 'Lavila'
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/Lavila_streaming_v2evaluator_deepseek_1_2.json'

# eval_model = 'EgoVLP'
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json'

# eval_model = 'CLIP'
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/CLIP_streaming_v2evaluator_deepseek_1_2.json'


parent_dir = os.path.dirname(eval_file)
eval_files = [os.path.join(parent_dir, f) for f in os.listdir(parent_dir) if f.startswith(eval_file.split('/')[-1])]
eval_result = []
for eval_file in eval_files:
    eval_result += load_multiple_json(eval_file)


task2number = {
    "Object Recognition": 0,
    "Attribute Perception": 0,
    "Text-Rich Understanding": 0,
    "Object Localization": 0,
    "Object State Change Recognition": 0,
    "Ego Object Localization": 0,
    "Ego Object State Change Recognition": 0,
    "Action Recognition": 0,
    "Object Function": 0,
    "Information Function": 0,
    "Action Reasoning": 0,
    "Task Understanding": 0,
    
}

task2score = {
    "Object Recognition": 0,
    "Attribute Perception": 0,
    "Text-Rich Understanding": 0,
    "Object Localization": 0,
    "Object State Change Recognition": 0,
    "Ego Object Localization": 0,
    "Ego Object State Change Recognition": 0,
    "Action Recognition": 0,
    "Object Function": 0,
    "Information Function": 0,
    "Action Reasoning": 0,
    "Task Understanding": 0,
    
    
}

task2recall = {
    "Object Recognition": 0,
    "Attribute Perception": 0,
    "Text-Rich Understanding": 0,
    "Object Localization": 0,
    "Object State Change Recognition": 0,
    "Ego Object Localization": 0,
    "Ego Object State Change Recognition": 0,
    "Action Recognition": 0,
    "Object Function": 0,
    "Information Function": 0,
    "Action Reasoning": 0,
    "Task Understanding": 0,
    
    
}

task2recall_score = {
    "Object Recognition": 0,
    "Attribute Perception": 0,
    "Text-Rich Understanding": 0,
    "Object Localization": 0,
    "Object State Change Recognition": 0,
    "Ego Object Localization": 0,
    "Ego Object State Change Recognition": 0,
    "Action Recognition": 0,
    "Object Function": 0,
    "Information Function": 0,
    "Action Reasoning": 0,
    "Task Understanding": 0,
    
    
}

task2nonrecall_pred = {
    "Object Recognition": 0,
    "Attribute Perception": 0,
    "Text-Rich Understanding": 0,
    "Object Localization": 0,
    "Object State Change Recognition": 0,
    "Ego Object Localization": 0,
    "Ego Object State Change Recognition": 0,
    "Action Recognition": 0,
    "Object Function": 0,
    "Information Function": 0,
    "Action Reasoning": 0,
    "Task Understanding": 0,
    
}

task2precision = {
    "Object Recognition": 0,
    "Attribute Perception": 0,
    "Text-Rich Understanding": 0,
    "Object Localization": 0,
    "Object State Change Recognition": 0,
    "Ego Object Localization": 0,
    "Ego Object State Change Recognition": 0,
    "Action Recognition": 0,
    "Object Function": 0,
    "Information Function": 0,
    "Action Reasoning": 0,
    "Task Understanding": 0,
}

def scoreMean(score):
    score = score.max(axis=0)
    return score.mean()

    
def validScoreMean(score):
    # FP
    nonrecall_pred = (score.max(axis=0) == 0).sum()
   
    # valid_score
    valid_score = np.zeros(score.shape[0])
    valid_recall = np.zeros(score.shape[0])
    valid_recall_score = np.zeros(score.shape[0])
    for i,s in enumerate(score):
        valid_s = s[s > 0]
        if len(valid_s) > 0:
            valid_score[i] = valid_s.sum() / (len(valid_s) + (nonrecall_pred / score.shape[0]))
            valid_score[i] = valid_s.sum() / (valid_s.sum() + (nonrecall_pred / score.shape[0]))
            # valid_score[i] = valid_s.max() / (1 + (nonrecall_pred / score.shape[0]))
            # valid_score[i] = valid_s.max() * len(valid_s) / (len(valid_s) + (nonrecall_pred / score.shape[0]))
            
            valid_recall[i] = 1
            valid_recall_score[i] = valid_s.sum() / len(valid_s)
    return valid_score.mean(), valid_recall.mean(), valid_recall_score.mean(), nonrecall_pred

BETA = 1
def validScoreF1(score):
    
    if score.shape[1] == 0:
        return 0, 0, 0, 0, 0
    # FP
    FP = (score.max(axis=0) == 0).sum()
    # TP
    TP = 0
   
    # valid_score
    valid_score = np.zeros(score.shape[0])
    valid_recall = np.zeros(score.shape[0])
    valid_recall_score = np.zeros(score.shape[0])
    for i,s in enumerate(score):
        valid_s = s[s > 0]
        if len(valid_s) > 0:
            # four type compute text-time precision
            valid_score[i] = valid_s.sum() / len(valid_s)
            # valid_score[i] = valid_s.max()
            # valid_score[i] = valid_s.sum() 
            
            valid_recall[i] = 1
            valid_recall_score[i] = valid_s.sum() / len(valid_s)
            # valid_recall_score[i] = valid_s.max()
            # valid_score[i] = valid_s.sum() 
    
    TP = valid_score.sum()
    precision = TP / (TP + FP)
    recall = valid_recall_score.mean()
    # breakpoint()
    if precision == 0 or recall == 0:
        F1 = 0
    else:
        F1 = (1 + BETA**2) * precision * recall / ((BETA**2 * precision) + recall)
    
    F1 = 2*TP / (2*TP + FP + score.shape[0] - valid_recall.sum())
    return F1, valid_recall.mean(), valid_recall_score.mean(), FP, precision

def topkValidScoreMean(score, k=10):
    score = score.max(axis=1)
    score = score[:k]
    valid_score = score[score > 0]
    return valid_score.mean()

total_fps = 0
total_kv_cache_size = 0
total_response_number = 0
total_answer_number = 0
total_precision = 0
for ll in eval_result:
    if eval_model in ll.keys():
        total_answer_number += len(ll['conversation'])
        this_turn_response_number = 0
        for response in ll[eval_model]:
            if response['role'].lower() == 'assistant':
                if 'fps' in response:
                    total_fps+=response['fps']
                total_response_number += 1
                this_turn_response_number += 1
                if 'kv_cache_size' in response:
                    total_kv_cache_size += response['kv_cache_size']
        task2number[ll['Task Type'].strip()] += 1
        # breakpoint()
        text_score = np.array(ll['evaluator_output_text']) / 10
        reponse_score = np.array(ll['evaluator_output_reponse']) / 10
        # if ll['Task Type'] == 'Ego Object State Change Recognition':
        if args.eval_mode == 'all' and eval_model not in ['EgoVLP', 'CLIP', 'Lavila']:
            score = (text_score+reponse_score)
        elif args.eval_mode == 'text':
            score = text_score
        elif args.eval_mode == 'response' or eval_model in ['EgoVLP', 'CLIP', 'Lavila']:
            score = reponse_score
        score_mean, recall_mean, recall_score_mean, nonrecall_pred, precision = validScoreF1(score)
        task2score[ll['Task Type'].strip()] += score_mean
        task2recall[ll['Task Type'].strip()] += recall_mean
        task2recall_score[ll['Task Type'].strip()] += recall_score_mean
        task2nonrecall_pred[ll['Task Type'].strip()] += nonrecall_pred
        task2precision[ll['Task Type'].strip()] += precision
        total_precision += nonrecall_pred / this_turn_response_number if this_turn_response_number > 0 else 0



for k,v in task2score.items():
    if task2number[k] == 0:
        task2score[k] = 0
        task2recall[k] = 0
        task2recall_score[k] = 0
        task2nonrecall_pred[k] = 0
        task2precision[k] = 0
    else:
        task2score[k] = v / task2number[k] * 100
        task2recall[k] = task2recall[k] / task2number[k] * 100
        task2recall_score[k] = task2recall_score[k] / task2number[k] * 100
        task2nonrecall_pred[k] = task2nonrecall_pred[k] / task2number[k]
        task2precision[k] = task2precision[k] / task2number[k] * 100
        
print(json.dumps(task2number, indent=4))
print(json.dumps({k: round(v, 2) for k,v in task2score.items()}, indent=4))
print(json.dumps(task2recall, indent=4))
# print(json.dumps(task2recall_score, indent=4))
# print(json.dumps(task2nonrecall_pred, indent=4))

print("Total question number: ", sum(task2number.values()))

print(f"Average Score: {sum(task2score.values())/len(task2number.values())}")

# Calculate average for first 8 task types
first_8_tasks = list(task2number.keys())[:8]
first_8_scores = [task2score[task] for task in first_8_tasks]
first_8_valid_scores = [score for score in first_8_scores if score > 0]
first_8_recall_scores = [task2recall[task] for task in first_8_tasks]
try:
    print(f"Average Score (First 8 Tasks): {sum(first_8_valid_scores)/len(first_8_valid_scores):.2f}")
except:
    print(f"Average Score (First 8 Tasks): 0.0")

# Calculate average for last 4 task types
last_4_tasks = list(task2number.keys())[-4:]
last_4_scores = [task2score[task] for task in last_4_tasks]
last_4_valid_scores = [score for score in last_4_scores if score > 0]
last_4_recall_scores = [task2recall[task] for task in last_4_tasks]
try:
    print(f"Average Score (Last 4 Tasks): {sum(last_4_valid_scores)/len(last_4_valid_scores):.2f}")
except:
    print(f"Average Score (Last 4 Tasks): 0.0")


print(f"Average Recall: {sum(task2recall.values())/len(task2number.values())}")
print(f"Average Recall Score: {sum(task2recall_score.values())/len(task2number.values())}")
print(f"Average Precision: {sum(task2precision.values())/len(task2number.values())}")


print(f"Total Question Number: {sum(task2number.values())}")
print(f"Average FPS: {total_fps/total_response_number}")
print(f"Average KV Cache: {total_kv_cache_size/total_response_number}")
print(f"Average Non-Recall Pred: {sum(task2nonrecall_pred.values())/sum(task2number.values())}")
print(f"Average Response Number: {total_response_number/sum(task2number.values())}")
print(f"Average Precision: {total_precision/sum(task2number.values()) * 100}")


print(f"[{sum(task2recall.values())/len(task2number.values()):.1f}, {sum(task2precision.values())/len(task2number.values()):.1f}]")

def generate_latex_table_row():
    # Generate LaTeX format table row
    print("\n# LaTeX format table row")
    print(f"& {eval_model} ", end="")

    # First 8 tasks
    for task in list(task2number.keys())[:8]:
        print(f"& {task2score[task]:.1f} ", end="")

    # Average of first 8 tasks
    print(f"& {sum(first_8_valid_scores)/len(first_8_valid_scores):.1f} ", end="")

    # Last 4 tasks
    for task in list(task2number.keys())[-4:]:
        print(f"& {task2score[task]:.1f} ", end="")

    # Average of last 4 tasks
    print(f"& {sum(last_4_valid_scores)/len(last_4_valid_scores):.1f} ", end="")

    # Overall average
    # print(f"& {sum(task2score.values())/len(task2number.values()):.1f} \\\\")

generate_latex_table_row()


# def generate_latex_table_row():
#     # Generate LaTeX format table row
#     print("\n# LaTeX format table row")
#     print(f"& {eval_model} ", end="")

#     # First 8 tasks
#     for task in list(task2number.keys())[:8]:
#         print(f"& {task2recall[task]:.1f} ", end="")

#     # Average of first 8 tasks
#     print(f"& {sum(first_8_recall_scores)/len(first_8_recall_scores):.1f} ", end="")

#     # Last 4 tasks
#     for task in list(task2number.keys())[-4:]:
#         print(f"& {task2recall[task]:.1f} ", end="")

#     # Average of last 4 tasks
#     print(f"& {sum(last_4_recall_scores)/len(last_4_recall_scores):.1f} ", end="")

#     # Overall average
#     # print(f"& {sum(task2score.values())/len(task2number.values()):.1f} \\\\")

# generate_latex_table_row()