dcase2024のメディアンフィルタの実装を見る

イベント予測でメディアンフィルタをどう使うのか

メディアンフィルタでどこからどこまでがイベントかを処理し,出力

細かいフレーム単位で判定

判定結果をフィルタで統合

欠点は以下

閾値を低くして感度を上げると,ノイズや静音を拾い,検出したイベントの時間が延長

逆に高くして精度を上げると,小さい音が拾えず,イベントの時間が短縮

実装を見にいく

code: python

class ClassWiseMedianFilter:

def __init__(self, filter_lens=(1, 1, 1)):

self.filter_lens = filter_lens

def __call__(self, x, **kwargs):

out = []

out.append(smoothed)

out = np.stack(out, -1)

return out

メディアンフィルタそのものではなく,それを使った場所を見にいく

code: python

batched_decode_preds(

filenames_strong,

self.encoder,

median_filter=self.median_filter,

thresholds=[],

)

strong_predsの内部構造が分からないが,おそらく各イベントのフレーム単位の予測をもらい,平滑化している?

code: python

def batched_decode_preds(

strong_preds,

filenames,

encoder,

median_filter=None,

pad_indx=None,

):

"""Decode a batch of predictions to dataframes. Each threshold gives a different dataframe and stored in a

dictionary

Args:

strong_preds: torch.Tensor, batch of strong predictions.

filenames: list, the list of filenames of the current batch.

encoder: ManyHotEncoder object, object used to decode predictions.

thresholds: list, the list of thresholds to be used for predictions.

median_filter: int, the number of frames for which to apply median window (smoothing).

pad_indx: list, the list of indexes which have been used for padding.

Returns:

dict of predictions, each keys is a threshold and the value is the DataFrame of predictions.

"""

# Init a dataframe per threshold

scores_raw = {}

scores_postprocessed = {}

prediction_dfs = {}

for threshold in thresholds:

filename = audio_id + ".wav"

if pad_indx is not None:

# pad_indxがリスト形式の場合とtensor形式の場合を対応

else:

c_scores = c_scores.transpose(0, 1).detach().cpu().numpy()

scores=c_scores,

timestamps=encoder._frame_to_time(np.arange(len(c_scores) + 1)),

event_classes=encoder.labels,

)

if median_filter is not None:

c_scores = median_filter(c_scores)

scores=c_scores,

timestamps=encoder._frame_to_time(np.arange(len(c_scores) + 1)),

event_classes=encoder.labels,

)

for c_th in thresholds:

pred = c_scores > c_th

pred = encoder.decode_strong(pred)

)

return scores_raw, scores_postprocessed, prediction_dfs