Skip to content

class PickScore

Bases: BaseMetric

Source code in aigve/metrics/text_video_alignment/similarity_based/pickscore/pick_infer.py 
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 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
@METRICS.register_module()
class PickScore(BaseMetric):
    """
    """
    def __init__(self,
                 model_name: str = "yuvalkirstain/PickScore_v1",
                 logit_scale: bool = False,
                 ) -> None:
        super().__init__()
        self.model_name = model_name
        self.logit_scale = logit_scale

        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        self.model =AutoModel.from_pretrained(self.model_name).eval().to(self.device)
        self.model.eval()


# def process(self, data_batch: dict, data_samples: Sequence[dict]) -> None:
    def process(self, data_batch: Sequence, data_samples: Sequence) -> None:
        """PickScore 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_prompts, input_videos = data_samples
        bsz = len(input_prompts)

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

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

        pickscore_sum, pickscore_cnt = 0, 0
        logit_scale = self.model.logit_scale.exp() if self.logit_scale else 1
        with torch.no_grad():
            for input_prompt, input_frames in zip(input_prompts, input_videos):

                input_prompt = input_prompt.to(self.device)
                text_feature = self.model.get_text_features(input_prompt)
                text_feature = text_feature / torch.norm(text_feature, dim=-1, keepdim=True)

                input_frames = input_frames.to(self.device)  # Add batch dimension and move the frame to the device
                frame_features = self.model.get_image_features(input_frames)
                frame_features = frame_features / torch.norm(frame_features, dim=-1, keepdim=True)

                pick_score = logit_scale *  (frame_features @ text_feature.T).mean().item()
                print('current pickscore', pick_score)
                pickscore_sum += pick_score
                pickscore_cnt += 1

        # get probabilities if you have multiple images to choose from
        # probs = torch.softmax(scores, dim=-1)
        pickscore_total_avg = pickscore_sum/pickscore_cnt
        result['pick_score'] = pickscore_total_avg

        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()

        pickscore_np = np.zeros(len(results))
        for i, result in enumerate(results):
            pickscore_np[i] = result['pick_score']

        pickscore_sim_mean = np.mean(pickscore_np) 

        print("Test results: PickScore={:.4f}"
              .format(pickscore_sim_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/similarity_based/pickscore/pick_infer.py 
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
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()

    pickscore_np = np.zeros(len(results))
    for i, result in enumerate(results):
        pickscore_np[i] = result['pick_score']

    pickscore_sim_mean = np.mean(pickscore_np) 

    print("Test results: PickScore={:.4f}"
          .format(pickscore_sim_mean))

    return result

process(data_batch, data_samples)

PickScore 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/similarity_based/pickscore/pick_infer.py 
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
def process(self, data_batch: Sequence, data_samples: Sequence) -> None:
    """PickScore 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_prompts, input_videos = data_samples
    bsz = len(input_prompts)

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

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

    pickscore_sum, pickscore_cnt = 0, 0
    logit_scale = self.model.logit_scale.exp() if self.logit_scale else 1
    with torch.no_grad():
        for input_prompt, input_frames in zip(input_prompts, input_videos):

            input_prompt = input_prompt.to(self.device)
            text_feature = self.model.get_text_features(input_prompt)
            text_feature = text_feature / torch.norm(text_feature, dim=-1, keepdim=True)

            input_frames = input_frames.to(self.device)  # Add batch dimension and move the frame to the device
            frame_features = self.model.get_image_features(input_frames)
            frame_features = frame_features / torch.norm(frame_features, dim=-1, keepdim=True)

            pick_score = logit_scale *  (frame_features @ text_feature.T).mean().item()
            print('current pickscore', pick_score)
            pickscore_sum += pick_score
            pickscore_cnt += 1

    # get probabilities if you have multiple images to choose from
    # probs = torch.softmax(scores, dim=-1)
    pickscore_total_avg = pickscore_sum/pickscore_cnt
    result['pick_score'] = pickscore_total_avg

    self.results.append(result)