def extract_pairwise_multi_position_embedding_nd(position_mat,
feat_dim,
wave_length=1000):
""" Extract multi-class position embedding
Args:
position_mat: [num_fg_classes, num_rois, num_rois, 4]
feat_dim: dimension of embedding feature
wave_length:
Returns:
embedding: [num_fg_classes, num_rois, num_rois, feat_dim]
"""
feat_range = nd.arange(0, feat_dim / 8)
dim_mat = nd.broadcast_power(lhs=nd.full((1, ), wave_length),
rhs=(8. / feat_dim) * feat_range)
dim_mat = nd.Reshape(dim_mat, shape=(1, 1, 1, 1, -1))
position_mat = nd.expand_dims(100.0 * position_mat, axis=4)
div_mat = nd.broadcast_div(lhs=position_mat, rhs=dim_mat)
sin_mat = nd.sin(data=div_mat)
cos_mat = nd.cos(data=div_mat)
# embedding, [num_fg_classes, num_rois, num_rois, 4, feat_dim/4]
embedding = nd.concat(sin_mat, cos_mat, dim=4)
embedding = nd.Reshape(embedding, shape=(0, 0, 0, feat_dim))
return embedding
def refine_bbox_nd(bbox, bbox_delta, im_info=None, means=None, stds=None):
xmin, ymin, xmax, ymax = nd.split(data=bbox, num_outputs=4, axis=1)
bbox_width = xmax - xmin + 1.
bbox_height = ymax - ymin + 1.
center_x = 0.5 * (xmin + xmax)
center_y = 0.5 * (ymin + ymax)
bbox_delta_reshape = nd.Reshape(data=bbox_delta, shape=(0, -1, 4))
dx, dy, dw, dh = nd.split(data=bbox_delta_reshape,
num_outputs=4,
axis=2,
squeeze_axis=1)
if (means is not None) and (stds is not None):
dx = dx * stds[0] + means[0]
dy = dy * stds[1] + means[1]
dw = dw * stds[2] + means[2]
dh = dh * stds[3] + means[3]
refine_center_x = nd.broadcast_add(lhs=center_x,
rhs=nd.broadcast_mul(lhs=bbox_width,
rhs=dx))
refine_center_y = nd.broadcast_add(lhs=center_y,
rhs=nd.broadcast_mul(lhs=bbox_height,
rhs=dy))
refined_width = nd.broadcast_mul(lhs=bbox_width, rhs=nd.exp(dw))
refined_height = nd.broadcast_mul(lhs=bbox_height, rhs=nd.exp(dh))
w_offset = 0.5 * (refined_width - 1.)
h_offset = 0.5 * (refined_height - 1.)
refined_xmin = nd.expand_dims(refine_center_x - w_offset, axis=1)
refined_ymin = nd.expand_dims(refine_center_y - h_offset, axis=1)
refined_xmax = nd.expand_dims(refine_center_x + w_offset, axis=1)
refined_ymax = nd.expand_dims(refine_center_y + h_offset, axis=1)
refined_bbox = nd.concat(refined_xmin,
refined_ymin,
refined_xmax,
refined_ymax,
dim=1)
if im_info is not None:
# assume im_info [[height, width, scale]] with shape (1,3)
im_hw = nd.slice_axis(im_info, axis=1, begin=0, end=2)
im_wh = nd.reverse(im_hw, axis=1)
im_wh = im_wh - 1.
im_wh = nd.tile(data=im_wh, reps=(1, 2))
im_wh = nd.Reshape(im_wh, shape=(1, 4, 1))
refined_bbox = nd.broadcast_minimum(lhs=refined_bbox, rhs=im_wh)
refined_bbox = nd.broadcast_maximum(lhs=refined_bbox,
rhs=nd.zeros_like(refined_bbox))
# print refined_bbox.debug_str()
return refined_bbox
def rnn(inputs, state, params):
# inputs和output都是num_steps个形状为(batch_size,vocab_size)
output = []
W_xh, W_hh, b_h, W_hq, b_q = params
H, = state # 只有第一个有参数H,见上,shape=(batch_size, num_hiddens)
for X in inputs: # 遍历num_steps个
H = nd.Reshape(nd.dot(X, W_xh) + nd.dot(H, W_hh) + b_h) # 计算隐藏状态,并作为返回值保存
Y = nd.dot(H, W_hq) + b_q
output.append(Y) # 追加
return output, (H, )
def extract_rank_embedding_nd(rank_dim, feat_dim, wave_length=1000):
rank_range = nd.arange(0, rank_dim)
feat_range = nd.arange(0, feat_dim / 2)
dim_mat = nd.broadcast_power(lhs=nd.full((1, ), wave_length),
rhs=(2. / feat_dim) * feat_range)
dim_mat = nd.Reshape(dim_mat, shape=(1, -1))
rank_mat = nd.expand_dims(rank_range, axis=1)
div_mat = nd.broadcast_div(lhs=rank_mat, rhs=dim_mat)
sin_mat = nd.sin(data=div_mat)
cos_mat = nd.cos(data=div_mat)
embedding = nd.concat(sin_mat, cos_mat, dim=1)
return embedding
def nms_attention_nd(roi_feat,
position_mat,
nms_pair_pos_fc1_1_weight,
nms_pair_pos_fc1_1_bias,
nms_query_1_weight,
nms_query_1_bias,
nms_key_1_weight,
nms_key_1_bias,
nms_linear_out_1_weight,
nms_linear_out_1_bias,
num_rois,
dim=(1024, 1024, 1024),
fc_dim=(64, 16),
feat_dim=1024,
group=16,
index=1):
""" Attetion module with vectorized version
Args:
roi_feat: [num_rois, num_fg_classes, feat_dim]
position_mat: [num_fg_classes, num_rois, num_rois, 4]
num_rois: number of rois
dim: key, query and linear_out dim
fc_dim:
feat_dim:
group:
index:
Returns:
output: [num_rois, num_fg_classes, fc_dim]
"""
dim_group = (dim[0] / group, dim[1] / group, dim[2] / group)
roi_feat = nd.transpose(roi_feat, axes=(1, 0, 2))
# roi_feat_reshape, [num_fg_classes*num_rois, feat_dim]
roi_feat_reshape = nd.Reshape(roi_feat, shape=(-3, -2))
# position_embedding, [num_fg_classes, num_rois, num_rois, fc_dim[0]]
position_embedding = extract_pairwise_multi_position_embedding_nd(
position_mat, fc_dim[0])
# [num_fg_classes * num_rois * num_rois, fc_dim[0]]
position_embedding_reshape = nd.Reshape(position_embedding,
shape=(-1, fc_dim[0]))
# position_feat_1, [num_fg_classes * num_rois * num_rois, fc_dim[1]]
position_feat_1 = nd.FullyConnected(name='nms_pair_pos_fc1_' + str(index),
data=position_embedding_reshape,
weight=nms_pair_pos_fc1_1_weight,
bias=nms_pair_pos_fc1_1_bias,
num_hidden=fc_dim[1])
# position_feat_1, [num_fg_classes, num_rois, num_rois, fc_dim[1]]
position_feat_1 = nd.Reshape(position_feat_1,
shape=(-1, num_rois, num_rois, fc_dim[1]))
aff_weight = nd.Activation(data=position_feat_1, act_type='relu')
# aff_weight, [num_fg_classes, fc_dim[1], num_rois, num_rois]
aff_weight = nd.transpose(aff_weight, axes=(0, 3, 1, 2))
####################### multi head in batch###########################
assert dim[0] == dim[1], 'Matrix multi requires the same dims!'
# q_data, [num_fg_classes * num_rois, dim[0]]
q_data = nd.FullyConnected(name='nms_query_' + str(index),
data=roi_feat_reshape,
weight=nms_query_1_weight,
bias=nms_query_1_bias,
num_hidden=dim[0])
# q_data, [num_fg_classes, num_rois, group, dim_group[0]]
q_data_batch = nd.Reshape(q_data,
shape=(-1, num_rois, group, dim_group[0]))
q_data_batch = nd.transpose(q_data_batch, axes=(0, 2, 1, 3))
# q_data_batch, [num_fg_classes * group, num_rois, dim_group[0]]
q_data_batch = nd.Reshape(q_data_batch, shape=(-3, -2))
k_data = nd.FullyConnected(name='nms_key_' + str(index),
data=roi_feat_reshape,
weight=nms_key_1_weight,
bias=nms_key_1_bias,
num_hidden=dim[1])
# k_data, [num_fg_classes, num_rois, group, dim_group[1]]
k_data_batch = nd.Reshape(k_data,
shape=(-1, num_rois, group, dim_group[1]))
k_data_batch = nd.transpose(k_data_batch, axes=(0, 2, 1, 3))
# k_data_batch, [num_fg_classes * group, num_rois, dim_group[1]]
k_data_batch = nd.Reshape(k_data_batch, shape=(-3, -2))
v_data = roi_feat
aff = nd.batch_dot(lhs=q_data_batch,
rhs=k_data_batch,
transpose_a=False,
transpose_b=True)
# aff_scale, [num_fg_classes * group, num_rois, num_rois]
aff_scale = (1.0 / math.sqrt(float(dim_group[1]))) * aff
assert fc_dim[1] == group, 'Check the dimensions in attention!'
# [num_fg_classes * fc_dim[1], num_rois, num_rois]
aff_weight_reshape = nd.Reshape(aff_weight, shape=(-3, -2))
# weighted_aff, [num_fg_classes * fc_dim[1], num_rois, num_rois]
weighted_aff = nd.log(nd.maximum(aff_weight_reshape, 1e-6)) + aff_scale
# aff_softmax, [num_fg_classes * fc_dim[1], num_rois, num_rois]
aff_softmax = nd.softmax(data=weighted_aff,
axis=2,
name='nms_softmax_' + str(index))
aff_softmax_reshape = nd.Reshape(aff_softmax,
shape=(-1, fc_dim[1] * num_rois, 0))
# output_t, [num_fg_classes, fc_dim[1] * num_rois, feat_dim]
output_t = nd.batch_dot(lhs=aff_softmax_reshape, rhs=v_data)
# output_t_reshape, [num_fg_classes, fc_dim[1], num_rois, feat_dim]
output_t_reshape = nd.Reshape(output_t,
shape=(-1, fc_dim[1], num_rois, feat_dim))
# output_t_reshape, [fc_dim[1], feat_dim, num_rois, num_fg_classes]
output_t_reshape = nd.transpose(output_t_reshape, axes=(1, 3, 2, 0))
# output_t_reshape, [1, fc_dim[1] * feat_dim, num_rois, num_fg_classes]
output_t_reshape = nd.Reshape(output_t_reshape,
shape=(1, fc_dim[1] * feat_dim, num_rois,
-1))
linear_out = nd.Convolution(name='nms_linear_out_' + str(index),
data=output_t_reshape,
weight=nms_linear_out_1_weight,
bias=nms_linear_out_1_bias,
kernel=(1, 1),
num_filter=dim[2],
num_group=fc_dim[1])
# [dim[2], num_rois, num_fg_classes]
linear_out_reshape = nd.Reshape(linear_out, shape=(dim[2], num_rois, -1))
# [num_rois, num_fg_classes, dim[2]]
output = nd.transpose(linear_out_reshape, axes=(1, 2, 0))
return output
def forward(self, is_train, req, in_data, out_data, aux):
nms_start_time = time.time()
#inputs
cls_score = in_data[0]
bbox_pred = in_data[1]
rois = in_data[2]
im_info = in_data[3]
fc_all_2_relu = in_data[4]
nms_rank_weight = in_data[5]
nms_rank_bias = in_data[6]
roi_feat_embedding_weight = in_data[7]
roi_feat_embedding_bias = in_data[8]
nms_pair_pos_fc1_1_weight = in_data[9]
nms_pair_pos_fc1_1_bias = in_data[10]
nms_query_1_weight = in_data[11]
nms_query_1_bias = in_data[12]
nms_key_1_weight = in_data[13]
nms_key_1_bias = in_data[14]
nms_linear_out_1_weight = in_data[15]
nms_linear_out_1_bias = in_data[16]
nms_logit_weight = in_data[17]
nms_logit_bias = in_data[18]
if self.has_non_gt_index:
non_gt_index = in_data[19]
else:
non_gt_index = None
if self.nongt_dim is not None:
cls_score_nongt = nd.slice_axis(data=cls_score,
axis=0,
begin=0,
end=self.nongt_dim)
# cls_score_nongt = monitor_wrapper(cls_score_nongt, 'cls_score_nongt')
bbox_pred_nongt = nd.slice_axis(data=bbox_pred,
axis=0,
begin=0,
end=self.nongt_dim)
elif non_gt_index is not None:
cls_score_nongt = nd.take(a=cls_score, indices=non_gt_index)
bbox_pred_nongt = nd.take(a=bbox_pred, indices=non_gt_index)
else:
cls_score_nongt = cls_score
bbox_pred_nongt = bbox_pred
bbox_pred_nongt = nd.BlockGrad(bbox_pred_nongt)
# remove batch idx and gt roi
sliced_rois = nd.slice_axis(data=rois, axis=1, begin=1, end=None)
if self.nongt_dim is not None:
sliced_rois = nd.slice_axis(data=sliced_rois,
axis=0,
begin=0,
end=self.nongt_dim)
elif non_gt_index is not None:
sliced_rois = nd.take(a=sliced_rois, indices=non_gt_index)
# bbox_pred_nobg, [num_rois, 4*(num_reg_classes-1)]
bbox_pred_nobg = nd.slice_axis(data=bbox_pred_nongt,
axis=1,
begin=4,
end=None)
# [num_boxes, 4, num_reg_classes-1]
refined_bbox = refine_bbox_nd(sliced_rois,
bbox_pred_nobg,
im_info,
means=self.bbox_means,
stds=self.bbox_stds)
# softmax cls_score to cls_prob, [num_rois, num_classes]
cls_prob = nd.softmax(data=cls_score_nongt, axis=-1)
cls_prob_nobg = nd.slice_axis(cls_prob, axis=1, begin=1, end=None)
sorted_cls_prob_nobg = nd.sort(data=cls_prob_nobg,
axis=0,
is_ascend=False)
# sorted_score, [first_n, num_fg_classes]
sorted_score = nd.slice_axis(sorted_cls_prob_nobg,
axis=0,
begin=0,
end=self.first_n,
name='sorted_score')
max_score_per_class = sorted_score.max(axis=0)
max_score_per_class_numpy = max_score_per_class.asnumpy()
valid_class_thresh = self.class_thresh
valid_class_thresh = np.minimum(valid_class_thresh,
max_score_per_class_numpy.max())
valid_class_indices = np.where(
max_score_per_class_numpy >= valid_class_thresh)[0]
invalid_class_indices = np.where(
max_score_per_class_numpy < valid_class_thresh)[0]
num_valid_classes = len(valid_class_indices)
valid_class_indices_nd = nd.array(valid_class_indices,
ctx=sorted_score.context)
# sort by score
rank_indices = nd.argsort(data=cls_prob_nobg, axis=0, is_ascend=False)
# first_rank_indices, [first_n, num_fg_classes]
first_rank_indices = nd.slice_axis(rank_indices,
axis=0,
begin=0,
end=self.first_n)
valid_first_rank_indices = first_rank_indices.transpose().take(
valid_class_indices_nd).transpose()
# sorted_bbox, [first_n, num_fg_classes, 4, num_reg_classes-1]
sorted_bbox = nd.take(a=refined_bbox, indices=first_rank_indices)
if self.class_agnostic:
# sorted_bbox, [first_n, num_fg_classes, 4]
sorted_bbox = nd.Reshape(sorted_bbox,
shape=(0, 0, 0),
name='sorted_bbox')
else:
cls_mask = nd.arange(0, self.num_fg_classes)
cls_mask = nd.Reshape(cls_mask, shape=(1, -1, 1))
cls_mask = nd.broadcast_to(cls_mask, shape=(self.first_n, 0, 4))
# sorted_bbox, [first_n, num_fg_classes, 4]
sorted_bbox = nd.pick(data=sorted_bbox,
name='sorted_bbox',
index=cls_mask,
axis=3)
valid_sorted_bbox = sorted_bbox.transpose(
(1, 0, 2)).take(valid_class_indices_nd).transpose((1, 0, 2))
# sorted_bbox = monitor_wrapper(sorted_bbox, 'sorted_bbox')
# nms_rank_embedding, [first_n, 1024]
nms_rank_embedding = extract_rank_embedding_nd(self.first_n, 1024)
# nms_rank_feat, [first_n, 1024]
nms_rank_feat = nd.FullyConnected(name='nms_rank',
data=nms_rank_embedding,
num_hidden=128,
weight=nms_rank_weight,
bias=nms_rank_bias)
# nms_position_matrix, [num_valid_classes, first_n, first_n, 4]
nms_position_matrix = extract_multi_position_matrix_nd(
valid_sorted_bbox)
# roi_feature_embedding, [num_rois, 1024]
# fc_all_2_relu = monitor_wrapper(fc_all_2_relu, 'fc_all_2_relu')
roi_feat_embedding = nd.FullyConnected(
name='roi_feat_embedding',
data=fc_all_2_relu,
num_hidden=128,
weight=roi_feat_embedding_weight,
bias=roi_feat_embedding_bias)
# sorted_roi_feat, [first_n, num_valid_classes, 128]
sorted_roi_feat = nd.take(a=roi_feat_embedding,
indices=valid_first_rank_indices)
# vectorized nms
# nms_embedding_feat, [first_n, num_valid_classes, 128]
nms_embedding_feat = nd.broadcast_add(lhs=sorted_roi_feat,
rhs=nd.expand_dims(nms_rank_feat,
axis=1))
# nms_attention_1, [first_n, num_valid_classes, 1024]
nms_attention_1 = nms_attention_nd(
nms_embedding_feat,
nms_position_matrix,
nms_pair_pos_fc1_1_weight,
nms_pair_pos_fc1_1_bias,
nms_query_1_weight,
nms_query_1_bias,
nms_key_1_weight,
nms_key_1_bias,
nms_linear_out_1_weight,
nms_linear_out_1_bias,
num_rois=self.first_n,
index=1,
group=self.nms_attention_group,
dim=self.nms_attention_dim,
fc_dim=self.nms_attention_fc_dim,
feat_dim=self.nms_attention_feat_dim)
nms_all_feat_1 = nms_embedding_feat + nms_attention_1
nms_all_feat_1_relu = nd.Activation(data=nms_all_feat_1,
act_type='relu',
name='nms_all_feat_1_relu')
# [first_n * num_valid_classes, 1024]
nms_all_feat_1_relu_reshape = nd.Reshape(nms_all_feat_1_relu,
shape=(-3, -2))
# logit, [first_n * num_valid_classes, num_thresh]
nms_conditional_logit = nd.FullyConnected(
name='nms_logit',
data=nms_all_feat_1_relu_reshape,
num_hidden=self.num_thresh,
weight=nms_logit_weight,
bias=nms_logit_bias)
# logit_reshape, [first_n, num_valid_classes, num_thresh]
nms_conditional_logit_reshape = nd.Reshape(nms_conditional_logit,
shape=(self.first_n,
num_valid_classes,
self.num_thresh))
nms_conditional_score = nd.Activation(
data=nms_conditional_logit_reshape,
act_type='sigmoid',
name='nms_conditional_score')
if num_valid_classes == self.num_fg_classes:
full_nms_conditional_score = nms_conditional_score
else:
full_nms_conditional_score = nd.concat(
nms_conditional_score,
nd.zeros(
(self.first_n, self.num_fg_classes - num_valid_classes,
self.num_thresh),
ctx=nms_conditional_score.context),
dim=1)
all_indexes = np.concatenate(
(valid_class_indices, invalid_class_indices))
restore_indexes = np.zeros((self.num_fg_classes))
restore_indexes[all_indexes] = np.arange(self.num_fg_classes)
restore_indexes = nd.array(restore_indexes,
ctx=nms_conditional_score.context)
full_nms_conditional_score = full_nms_conditional_score.transpose(
(1, 0, 2)).take(restore_indexes).transpose((1, 0, 2))
sorted_score_reshape = nd.expand_dims(sorted_score, axis=2)
# sorted_score_reshape = nd.BlockGrad(sorted_score_reshape)
nms_multi_score = nd.broadcast_mul(lhs=sorted_score_reshape,
rhs=full_nms_conditional_score)
_ = nms_multi_score.mean().asnumpy()
all_time = time.time() - nms_start_time
if 'learn_nms_time' not in globals().keys(
) or 'learn_nms_count' not in globals().keys():
globals()['learn_nms_time'] = []
globals()['learn_nms_count'] = 0
if globals()['learn_nms_count'] >= 1000:
globals()['learn_nms_time'].pop(0)
globals()['learn_nms_time'].append(all_time)
else:
globals()['learn_nms_time'].append(all_time)
globals()['learn_nms_count'] += 1
if globals()['learn_nms_count'] % 250 == 0:
print("--->> learn nms running average time cost: {}".format(
float(sum(globals()['learn_nms_time'])) /
(1000 if globals()['learn_nms_count'] > 1000 else
globals()['learn_nms_count'])))
self.assign(out_data[0], req[0], nms_multi_score)
self.assign(out_data[1], req[1], sorted_bbox)
self.assign(out_data[2], req[2], sorted_score)