def default_aggregation_func(score_arr):
"""
This is the default function for make video-level prediction
:param score_arr: a 3-dim array with (frame, crop, class) layout
:return:
"""
return softmax(score_arr.max(axis=1).mean(axis=0))
def sliding_window_aggregation_func(score, spans=[1, 2, 4, 8, 16], overlap=0.2, norm=True, fps=1):
"""
This is the aggregation function used for ActivityNet Challenge 2016
:param score:
:param spans:
:param overlap:
:param norm:
:param fps:
:return:
"""
frm_max = score.max(axis=1)
slide_score = []
def top_k_pool(scores, k):
return np.sort(scores, axis=0)[-k:, :].mean(axis=0)
for t_span in spans:
span = t_span * fps
step = int(np.ceil(span * (1-overlap)))
local_agg = [frm_max[i: i+span].max(axis=0) for i in xrange(0, frm_max.shape[0], step)]
k = max(15, len(local_agg)/4)
slide_score.append(top_k_pool(np.array(local_agg), k))
out_score = np.mean(slide_score, axis=0)
if norm:
return softmax(out_score)
else:
return out_score
def sliding_window_aggregation_func(score, spans=[1, 2, 4, 8, 16], overlap=0.2, norm=True, fps=1):
"""
This is the aggregation function used for ActivityNet Challenge 2016
:param score:
:param spans:
:param overlap:
:param norm:
:param fps:
:return:
"""
frm_max = score.max(axis=1)
slide_score = []
def top_k_pool(scores, k):
return np.sort(scores, axis=0)[-k:, :].mean(axis=0)
for t_span in spans:
span = t_span * fps
step = int(np.ceil(span * (1 - overlap)))
local_agg = [frm_max[i : i + span].max(axis=0) for i in xrange(0, frm_max.shape[0], step)]
k = max(15, len(local_agg) / 4)
slide_score.append(top_k_pool(np.array(local_agg), k))
out_score = np.mean(slide_score, axis=0)
if norm:
return softmax(out_score)
else:
return out_score
def default_aggregation_func(score_arr):
"""
This is the default function for make video-level prediction
:param score_arr: a 3-dim array with (frame, crop, class) layout
:return:
"""
return softmax(score_arr.max(axis=1).mean(axis=0))
def sliding_window_aggregation_func(score,
spans=[1, 2, 4, 8, 16],
overlap=0.2,
norm=True,
fps=1):
"""
Aggregate scores from models
:param score: (#frames, #models, #classes)
:return: (#classes)
The basica idea is to perform max pooling across models, and smooth the scores via sliding window across along frames, and finally average the top-k resutls across smoothed frames.
"""
frm_max = score.max(axis=1)
slide_score = []
def top_k_pool(scores, k):
return np.sort(scores, axis=0)[-k:, :].mean(axis=0)
for t_span in spans:
span = t_span * fps
step = int(np.ceil(span * (1 - overlap)))
local_agg = [
frm_max[i:i + span].max(axis=0)
for i in xrange(0, frm_max.shape[0], step)
]
k = max(15, len(local_agg) / 4)
slide_score.append(top_k_pool(np.array(local_agg), k))
out_score = np.mean(slide_score, axis=0)
if norm:
return softmax(out_score)
else:
return out_score
def top_k_aggregation_func(score_arr, k, normalization=True, crop_agg=None):
crop_agg = np.mean if crop_agg is None else crop_agg
if normalization:
return softmax(
np.sort(crop_agg(score_arr, axis=1), axis=0)[-k:, :].mean(axis=0))
else:
return np.sort(crop_agg(score_arr, axis=1),
axis=0)[-k:, :].mean(axis=0)
def default_fusion_func(major_score, other_scores, fusion_weights, norm=True):
assert len(other_scores) == len(fusion_weights)
out_score = major_score
for s, w in zip(other_scores, fusion_weights):
out_score += s * w
if norm:
return softmax(out_score)
else:
return out_score
def default_fusion_func(major_score, other_scores, fusion_weights, norm=True):
assert len(other_scores) == len(fusion_weights)
out_score = major_score
for s, w in zip(other_scores, fusion_weights):
out_score += s * w
if norm:
return softmax(out_score)
else:
return out_score
def sci_crop_stream_aggregation_func(score_arr, score_arr_1, normalization=True, crop_agg=None):
a,b,c = score_arr.shape[0:3]
frame_score = []
frame_score_1 = []
for i in range(a):
stream_sum = 0
stream_sum_1 = 0
scipp_sum = 0
scipp_sum_1 = 0
for j in range(b):
stream = score_arr[i][j]
stream_1 = score_arr_1[i][j]
stream_sum += sci_index(stream)
stream_sum_1 += sci_index(stream_1)
scipp_sum += sci_index(stream) * softmax(stream)
scipp_sum_1 += sci_index(stream) * softmax(stream_1)
p1 = scipp_sum / stream_sum
p2 = scipp_sum_1 / stream_sum_1
# p = sci_index(p1) * softmax(p1) + sci_index(p2) * softmax(p2)/(sci_index(p1) + sci_index(p2))
frame_score.append(p1)
frame_score_1.append(p2)
frame_score = np.array(frame_score)
frame_score_1 = np.array(frame_score_1)
frame_score = (frame_score).max(axis=0)
frame_score_1 = (frame_score_1).max(axis=0)
frame_score = softmax(frame_score)
frame_score_1 = softmax(frame_score_1)
frame_score = sci_index(frame_score, normalization=False) * frame_score + sci_index(frame_score_1, normalization=False) * frame_score_1
# frame_score += frame_score_1
#frame_score = (frame_score).max(axis=0)
if normalization:
return softmax((frame_score) )
else:
return (frame_score)
def default_aggregation_func(score_arr, normalization=True, crop_agg=None):
"""
This is the default function for make video-level prediction
:param score_arr: a 3-dim array with (frame, crop, class) layout
:return:
"""
crop_agg = np.mean if crop_agg is None else crop_agg
if normalization:
return softmax(crop_agg(score_arr, axis=1).mean(axis=0))
else:
return crop_agg(score_arr, axis=1).mean(axis=0)
def default_aggregation_func(score_arr, normalization=True, crop_agg=None):
"""
This is the default function for make video-level prediction
:param score_arr: a 3-dim array with (frame, crop, class) layout
:return:
"""
crop_agg = np.mean if crop_agg is None else crop_agg
if normalization:
return softmax(crop_agg(score_arr, axis=1).mean(axis=0))
else:
return crop_agg(score_arr, axis=1).mean(axis=0)
def default_aggregation_func(score_arr, normalization=False, crop_agg=None):
"""
This is the default function for make video-level prediction
:param score_arr: a 3-dim array with (frame, crop, class) layout
:return:
"""
crop_agg = np.mean if crop_agg is None else crop_agg
if normalization:
a = softmax(crop_agg(score_arr, axis=1).mean(axis=0))
return a
else:
a = crop_agg(score_arr, axis=1)
a = a.mean(axis=0)
a = a.reshape(1,1,-1)
return a
def sci_index(x, normalization = True):
if normalization:
x = softmax(x)
scip = (101 * np.max(x)/np.sum(x) - 1)/(101 - 1)
return scip
def top_k_aggregation_func(score_arr, k, normalization=True, crop_agg=None):
crop_agg = np.mean if crop_agg is None else crop_agg
if normalization:
return softmax(np.sort(crop_agg(score_arr, axis=1), axis=0)[-k:, :].mean(axis=0))
else:
return np.sort(crop_agg(score_arr, axis=1), axis=0)[-k:, :].mean(axis=0)