utils package¶
Submodules¶
utils.adj_matrix module¶
- utils.adj_matrix.Norm_F(feature, norm_flag)¶
- Args:
feature: features to be normalized, (b,c,1,n) norm_flag: normalization flag: constant, l1norm, l2norm
- Returns:
normalized feature in the third dimension size: (b,c,1,n)
- utils.adj_matrix.build_idx(num)¶
- Args:
num: number of graph nodes
- Returns:
indices of each nodes
- utils.adj_matrix.buildnear(i, j, len, edge)¶
- utils.adj_matrix.gen_adj(block_width, batch)¶
- Args:
block_width: feature size width/height used for spatial GCN batch: batchsize
- Returns:
adj matrices
- utils.adj_matrix.gen_adj2(num_block)¶
- utils.adj_matrix.gen_adj_coo(block_width, batch)¶
generate adjacent matrices by COO form
- utils.adj_matrix.gen_adj_nearst(num_block, part_features)¶
enable nearst connections for the graph embedding Args:
num_block: part_features:
Returns:
- utils.adj_matrix.gen_adj_sim(num_block, part_features1, part_features2, enable_mask=False)¶
default similarity function, calculate similarity between two input features
enable_mask: encourage sparse connections when similarities higher than the threshold
- Args:
num_block: abandoned feature part_features1: vector1, size = b,c,n1 part_features2: vector2, size = b,c,n2 enable_mask: True/ False
- Returns:
similarity matrices: size: n1 * n2
- utils.adj_matrix.gen_adj_sim_old(num_block, part_features)¶
- utils.adj_matrix.gen_adj_topk(num_block, part_features)¶
- utils.adj_matrix.get_sim_cross(edge)¶
handcrafted similarity matrices
- Args:
edge:
Returns:
- utils.adj_matrix.get_sim_cross2(edge)¶
handcrafted similarity matrices
- utils.adj_matrix.get_sim_local(edge1, edge2)¶
get local connections Args:
edge1: edge2:
Returns:
- utils.adj_matrix.mul_dis(feature1, feature2)¶
matmul similarity function
- utils.adj_matrix.sim_dis(feature1, feature2)¶
cosine simlarity function
utils.meter module¶
utils.metrics module¶
- class utils.metrics.R1_mAP(num_query, max_rank=50, feat_norm=True, method='euclidean', reranking=False)¶
Bases:
object
- compute()¶
- compute_acc()¶
- reset()¶
- update(output)¶
- utils.metrics.cosine_similarity(qf, gf)¶
- utils.metrics.euclidean_distance(qf, gf)¶
- utils.metrics.eval_func(distmat, q_pids, g_pids, max_rank=50)¶
Evaluation with market1501 metric Key: for each query identity, its gallery images from the same camera view are discarded.
utils.reranking module¶
Created on Fri, 25 May 2018 20:29:09
@author: luohao
- utils.reranking.re_ranking(probFea, galFea, k1, k2, lambda_value, local_distmat=None, only_local=False)¶