Skip to content

class DSGScore

Bases: BaseMetric

Source code in aigve/metrics/text_video_alignment/gpt_based/dsg/dsg_eval.py 
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
@METRICS.register_module()
class DSGScore(BaseMetric):
    '''
    '''
    def __init__(self, 
                 vqa_model_name: str = "InstructBLIP",
                 verbose: bool = False):
        super().__init__()

        self.vqa_model_name = vqa_model_name
        assert self.vqa_model_name in ["InstructBLIP", "MPLUG"]
        if self.vqa_model_name == 'InstructBLIP':
            self.vqa_model = InstructBLIP()
        else:
            self.vqa_model = MPLUG()

        self.verbose = verbose
        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")


    def evaluate_image_dsg(self, qid_list, frame_index, frame):
        '''evaluate a generated image with DSG
        '''
        if self.verbose:
            print("#"*50)
            print("2) Answer questions given the generated image, with VQA")
            print("#"*50)

        # 2) answer questions with the generated image
        qid2answer = {}
        qid2scores = {}

        qid2tuple, qid2dependency, qid2question = qid_list
        for id, question in qid2question.items():
            answer = self.vqa_model.vqa(image=frame, question=question)
            print(answer)
            qid2answer[id] = answer
            qid2scores[id] = float('yes' in answer)

        average_score_without_dep = sum(qid2scores.values()) / len(qid2scores)
        print(average_score_without_dep, qid2answer, qid2scores)

        if self.verbose:
            print("#"*50)
            print("3) Zero-out scores from invalid questions")
            print("#"*50)

        # 3) zero-out scores from invalid questions 
        qid2validity = {}
        qid2scores_after_filtering = deepcopy(qid2scores)

        # print('qid2scores', qid2scores)
        # print('qid2dependency', qid2dependency)
        for id, parent_ids in qid2dependency.items():
            # zero-out scores if parent questions are answered 'no'
            any_parent_answered_no = False
            for parent_id in parent_ids:
                parent_id = list(parent_id)[0]
                if parent_id == 0:
                    continue
                if qid2scores[parent_id] == 0:
                    any_parent_answered_no = True
                    break
            if any_parent_answered_no:
                qid2scores_after_filtering[id] = 0.0
                qid2validity[id] = False
            else:
                qid2validity[id] = True

        if self.verbose:
            print("Per-quesiton eval results (after using dependency)")
            for id in qid2question:
                print("ID", id)
                print("question", qid2question[id])
                print("answer", qid2answer[id])
                print("validity", qid2validity[id])
                print("score (before filtering)", qid2scores[id])
                print("score (after filtering)", qid2scores_after_filtering[id])
                print()

        if self.verbose:
            print("#"*50)
            print("4) Calculate the final score by averaging")
            print("#"*50)

        average_score_with_dep = sum(qid2scores_after_filtering.values()) / len(qid2scores)

        return {
            'frame_index': frame_index,
            'qid2tuple': qid2tuple,
            'qid2dependency': qid2dependency,
            'qid2question': qid2question,
            'qid2answer': qid2answer,
            'qid2scores': qid2scores,
            'qid2validity': qid2validity,
            'average_score_with_dependency': average_score_with_dep,
            'average_score_without_dependency': average_score_without_dep
        }


    def process(self, data_batch: Sequence, data_samples: Sequence) -> None:
        """DSGScore process
        Process one batch of data samples and predictions. The processed
        results should be stored in ``self.results``, which will be used to
        compute the metrics when all batches have been processed.

        Args:
            data_batch (Sequence): A batch of data from the dataloader.
            data_samples (Sequence): A batch of data samples that
                contain annotations and predictions.
        """

        result = dict()

        input_qid_lists, input_videos = data_samples
        bsz = len(input_qid_lists)
        # print('input_qid_lists: ', input_qid_lists)

        # Ensure prompt_input is a tensor
        if isinstance(input_qid_lists, tuple):
            input_qid_lists = list(input_qid_lists)

        if isinstance(input_videos, tuple):
            input_videos = list(input_videos)

        average_dep_score_list, average_wo_dep_score_list = [], []
        for input_qid_list, input_video in zip([input_qid_lists], input_videos):
            evaluate_dict_list = []
            dep_score, wo_dep_score = [], []
            for index, frame in enumerate(input_video):
                # print('input_qid_list: ', input_qid_list)
                evaluate_dict = self.evaluate_image_dsg(qid_list=input_qid_list, 
                                                        frame_index=index, 
                                                        frame=frame)
                evaluate_dict_list.append(evaluate_dict)
                frame_average_score_with_dependency = evaluate_dict['average_score_with_dependency']
                dep_score.append(frame_average_score_with_dependency)
                frame_average_score_without_dependency = evaluate_dict['average_score_without_dependency']
                wo_dep_score.append(frame_average_score_without_dependency)
            avg_dep_score, avg_wo_dep_score = sum(dep_score)/len(dep_score), sum(wo_dep_score)/len(dep_score)
            average_dep_score_list.append(avg_dep_score)
            average_wo_dep_score_list.append(avg_wo_dep_score)


        result['average_dep_dgs_score'] = sum(average_dep_score_list)/len(average_dep_score_list)
        result['average_wo_dep_dgs_score'] = sum(average_wo_dep_score_list)/len(average_wo_dep_score_list)

        self.results.append(result)


    def compute_metrics(self, results: list) -> Dict[str, float]:
        """Compute the metrics from processed results.

        Args:
            results (list): The processed results of each batch.

        Returns:
            Dict[str, float]: The computed metrics. The keys are the names of
            the metrics, and the values are corresponding results.
        """
        logger: MMLogger = MMLogger.get_current_instance()

        dep_dsg_score_np = np.zeros(len(results))
        wo_dep_dsg_score_np = np.zeros(len(results))
        for i, result in enumerate(results):
            dep_dsg_score_np[i] = result['average_dep_dgs_score']
            wo_dep_dsg_score_np[i] = result['average_wo_dep_dgs_score']

        dep_dsg_score_np_mean = np.mean(dep_dsg_score_np) 
        wo_dep_dsg_score_np_mean = np.mean(wo_dep_dsg_score_np)

        print("Test results: dsg score with dependency={:.4f}"
              .format(dep_dsg_score_np_mean))
        print("Test results: dsg score without dependency={:.4f}"
              .format(wo_dep_dsg_score_np_mean))

        return result

compute_metrics(results)

Compute the metrics from processed results.

Parameters:

Name Type Description Default
results list

The processed results of each batch.

 required

Returns:

Type Description
Dict[str, float]

Dict[str, float]: The computed metrics. The keys are the names of
Dict[str, float]

the metrics, and the values are corresponding results.
Source code in aigve/metrics/text_video_alignment/gpt_based/dsg/dsg_eval.py 
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
def compute_metrics(self, results: list) -> Dict[str, float]:
    """Compute the metrics from processed results.

    Args:
        results (list): The processed results of each batch.

    Returns:
        Dict[str, float]: The computed metrics. The keys are the names of
        the metrics, and the values are corresponding results.
    """
    logger: MMLogger = MMLogger.get_current_instance()

    dep_dsg_score_np = np.zeros(len(results))
    wo_dep_dsg_score_np = np.zeros(len(results))
    for i, result in enumerate(results):
        dep_dsg_score_np[i] = result['average_dep_dgs_score']
        wo_dep_dsg_score_np[i] = result['average_wo_dep_dgs_score']

    dep_dsg_score_np_mean = np.mean(dep_dsg_score_np) 
    wo_dep_dsg_score_np_mean = np.mean(wo_dep_dsg_score_np)

    print("Test results: dsg score with dependency={:.4f}"
          .format(dep_dsg_score_np_mean))
    print("Test results: dsg score without dependency={:.4f}"
          .format(wo_dep_dsg_score_np_mean))

    return result

evaluate_image_dsg(qid_list, frame_index, frame)

evaluate a generated image with DSG

Source code in aigve/metrics/text_video_alignment/gpt_based/dsg/dsg_eval.py 
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
def evaluate_image_dsg(self, qid_list, frame_index, frame):
    '''evaluate a generated image with DSG
    '''
    if self.verbose:
        print("#"*50)
        print("2) Answer questions given the generated image, with VQA")
        print("#"*50)

    # 2) answer questions with the generated image
    qid2answer = {}
    qid2scores = {}

    qid2tuple, qid2dependency, qid2question = qid_list
    for id, question in qid2question.items():
        answer = self.vqa_model.vqa(image=frame, question=question)
        print(answer)
        qid2answer[id] = answer
        qid2scores[id] = float('yes' in answer)

    average_score_without_dep = sum(qid2scores.values()) / len(qid2scores)
    print(average_score_without_dep, qid2answer, qid2scores)

    if self.verbose:
        print("#"*50)
        print("3) Zero-out scores from invalid questions")
        print("#"*50)

    # 3) zero-out scores from invalid questions 
    qid2validity = {}
    qid2scores_after_filtering = deepcopy(qid2scores)

    # print('qid2scores', qid2scores)
    # print('qid2dependency', qid2dependency)
    for id, parent_ids in qid2dependency.items():
        # zero-out scores if parent questions are answered 'no'
        any_parent_answered_no = False
        for parent_id in parent_ids:
            parent_id = list(parent_id)[0]
            if parent_id == 0:
                continue
            if qid2scores[parent_id] == 0:
                any_parent_answered_no = True
                break
        if any_parent_answered_no:
            qid2scores_after_filtering[id] = 0.0
            qid2validity[id] = False
        else:
            qid2validity[id] = True

    if self.verbose:
        print("Per-quesiton eval results (after using dependency)")
        for id in qid2question:
            print("ID", id)
            print("question", qid2question[id])
            print("answer", qid2answer[id])
            print("validity", qid2validity[id])
            print("score (before filtering)", qid2scores[id])
            print("score (after filtering)", qid2scores_after_filtering[id])
            print()

    if self.verbose:
        print("#"*50)
        print("4) Calculate the final score by averaging")
        print("#"*50)

    average_score_with_dep = sum(qid2scores_after_filtering.values()) / len(qid2scores)

    return {
        'frame_index': frame_index,
        'qid2tuple': qid2tuple,
        'qid2dependency': qid2dependency,
        'qid2question': qid2question,
        'qid2answer': qid2answer,
        'qid2scores': qid2scores,
        'qid2validity': qid2validity,
        'average_score_with_dependency': average_score_with_dep,
        'average_score_without_dependency': average_score_without_dep
    }

process(data_batch, data_samples)

DSGScore process Process one batch of data samples and predictions. The processed results should be stored in self.results, which will be used to compute the metrics when all batches have been processed.

Parameters:

Name Type Description Default
data_batch Sequence

A batch of data from the dataloader.

 required
data_samples Sequence

A batch of data samples that contain annotations and predictions.

 required
Source code in aigve/metrics/text_video_alignment/gpt_based/dsg/dsg_eval.py 
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
def process(self, data_batch: Sequence, data_samples: Sequence) -> None:
    """DSGScore process
    Process one batch of data samples and predictions. The processed
    results should be stored in ``self.results``, which will be used to
    compute the metrics when all batches have been processed.

    Args:
        data_batch (Sequence): A batch of data from the dataloader.
        data_samples (Sequence): A batch of data samples that
            contain annotations and predictions.
    """

    result = dict()

    input_qid_lists, input_videos = data_samples
    bsz = len(input_qid_lists)
    # print('input_qid_lists: ', input_qid_lists)

    # Ensure prompt_input is a tensor
    if isinstance(input_qid_lists, tuple):
        input_qid_lists = list(input_qid_lists)

    if isinstance(input_videos, tuple):
        input_videos = list(input_videos)

    average_dep_score_list, average_wo_dep_score_list = [], []
    for input_qid_list, input_video in zip([input_qid_lists], input_videos):
        evaluate_dict_list = []
        dep_score, wo_dep_score = [], []
        for index, frame in enumerate(input_video):
            # print('input_qid_list: ', input_qid_list)
            evaluate_dict = self.evaluate_image_dsg(qid_list=input_qid_list, 
                                                    frame_index=index, 
                                                    frame=frame)
            evaluate_dict_list.append(evaluate_dict)
            frame_average_score_with_dependency = evaluate_dict['average_score_with_dependency']
            dep_score.append(frame_average_score_with_dependency)
            frame_average_score_without_dependency = evaluate_dict['average_score_without_dependency']
            wo_dep_score.append(frame_average_score_without_dependency)
        avg_dep_score, avg_wo_dep_score = sum(dep_score)/len(dep_score), sum(wo_dep_score)/len(dep_score)
        average_dep_score_list.append(avg_dep_score)
        average_wo_dep_score_list.append(avg_wo_dep_score)


    result['average_dep_dgs_score'] = sum(average_dep_score_list)/len(average_dep_score_list)
    result['average_wo_dep_dgs_score'] = sum(average_wo_dep_score_list)/len(average_wo_dep_score_list)

    self.results.append(result)