"""
Based on code from:
@author: liaoxingyu
@contact: sherlockliao01@gmail.com
Adapted and extended by:
@author: mikwieczorek
"""
import torch
import torch.nn.functional as F
from torch import nn
def normalize(x, axis=-1):
"""Normalizing to unit length along the specified dimension.
Args:
x: pytorch Variable
Returns:
x: pytorch Variable, same shape as input
"""
x = 1. * x / (torch.norm(x, 2, axis, keepdim=True).expand_as(x) + 1e-12)
return x
def euclidean_dist(x, y):
"""
Args:
x: pytorch Variable, with shape [m, d]
y: pytorch Variable, with shape [n, d]
Returns:
dist: pytorch Variable, with shape [m, n]
"""
m, n = x.size(0), y.size(0)
xx = torch.pow(x, 2).sum(1, keepdim=True).expand(m, n)
yy = torch.pow(y, 2).sum(1, keepdim=True).expand(n, m).t()
dist = xx + yy
dist.addmm_(1, -2, x.float(), y.float().t())
dist = dist.clamp(min=1e-12).sqrt()
return dist
def cosine_similarity(x:torch.Tensor,y:torch.Tensor, eps:float=1e-12) -> torch.Tensor:
"""
Computes cosine similarity between two tensors.
Value == 1 means the same vector
Value == 0 means perpendicular vectors
"""
x_n, y_n = x.norm(dim=1)[:, None], y.norm(dim=1)[:, None]
x_norm = x / torch.max(x_n, eps * torch.ones_like(x_n))
y_norm = y / torch.max(y_n, eps * torch.ones_like(y_n))
sim_mt = torch.mm(x_norm, y_norm.transpose(0, 1))
return sim_mt
def cosine_dist(x:torch.Tensor,y:torch.Tensor, eps:float=1e-12) -> torch.Tensor:
"""
Computes cosine distance between two tensors.
The cosine distance is the inverse cosine similarity
-> cosine_distance = abs(-cosine_distance) to make it
similar in behaviour to euclidean distance
"""
sim_mt = cosine_similarity(x,y, eps)
return torch.abs(1-sim_mt).clamp(min=eps)
def hard_example_mining(dist_mat, labels, return_inds=False):
"""For each anchor, find the hardest positive and negative sample.
Args:
dist_mat: pytorch Variable, pair wise distance between samples, shape [N, N]
labels: pytorch LongTensor, with shape [N]
return_inds: whether to return the indices. Save time if `False`(?)
Returns:
dist_ap: pytorch Variable, distance(anchor, positive); shape [N]
dist_an: pytorch Variable, distance(anchor, negative); shape [N]
p_inds: pytorch LongTensor, with shape [N];
indices of selected hard positive samples; 0 <= p_inds[i] <= N - 1
n_inds: pytorch LongTensor, with shape [N];
indices of selected hard negative samples; 0 <= n_inds[i] <= N - 1
NOTE: Only consider the case in which all labels have same num of samples,
thus we can cope with all anchors in parallel.
"""
assert len(dist_mat.size()) == 2
assert dist_mat.size(0) == dist_mat.size(1)
N = dist_mat.size(0)
is_pos = labels.expand(N, N).eq(labels.expand(N, N).t())
is_neg = labels.expand(N, N).ne(labels.expand(N, N).t())
dist_ap, relative_p_inds = torch.max(
dist_mat[is_pos].contiguous().view(N, -1), 1, keepdim=True)
dist_an, relative_n_inds = torch.min(dist_mat[is_neg].contiguous().view(N, -1), 1, keepdim=True)
dist_ap = dist_ap.squeeze(1)
dist_an = dist_an.squeeze(1)
if return_inds:
ind = (labels.new().resize_as_(labels)
.copy_(torch.arange(0, N).long())
.unsqueeze(0).expand(N, N))
p_inds = torch.gather(
ind[is_pos].contiguous().view(N, -1), 1, relative_p_inds.data)
n_inds = torch.gather(
ind[is_neg].contiguous().view(N, -1), 1, relative_n_inds.data)
p_inds = p_inds.squeeze(1)
n_inds = n_inds.squeeze(1)
return dist_ap, dist_an, p_inds, n_inds
return dist_ap, dist_an
class TripletLoss(object):
"""Modified from Tong Xiao's open-reid (https://github.com/Cysu/open-reid).
Related Triplet Loss theory can be found in paper 'In Defense of the Triplet
Loss for Person Re-Identification'."""
def __init__(self, margin=None, dist_func='euclidean'):
self.margin = margin
if margin is not None:
self.ranking_loss = nn.MarginRankingLoss(margin=margin)
else:
self.ranking_loss = nn.SoftMarginLoss()
if dist_func == 'cosine':
self.dist_func = cosine_dist
elif dist_func == 'euclidean':
self.dist_func = euclidean_dist
def __call__(self, global_feat, labels, warmup_margin=False, print_data=False, normalize_feature=False, mask=None):
if normalize_feature:
global_feat = normalize(global_feat, axis=-1)
dist_mat = self.dist_func(global_feat, global_feat)
dist_ap, dist_an = hard_example_mining(
dist_mat, labels)
y = dist_an.new().resize_as_(dist_an).fill_(1)
if mask is not None:
dist_ap = dist_ap[mask]
dist_an = dist_an[mask]
y = y[mask]
if self.margin is not None:
loss = self.ranking_loss(dist_an, dist_ap, y)
else:
loss = self.ranking_loss(dist_an - dist_ap, y)
if print_data:
print(f'LOSS: {loss.item()}')
prec = (dist_an > dist_ap).data.float().mean()
print(f'precision: {prec}')
sm = (dist_an > dist_ap + self.margin).data.float().mean()
print(f'proportion of triplets that satisfy margin: {sm}')
d_ap = dist_ap.data.mean()
print(f'AP mean distance: {d_ap}')
d_an = dist_an.data.mean()
print(f'AN mean distance: {d_an}')
return loss, dist_ap, dist_an
class CrossEntropyLabelSmooth(nn.Module):
"""Cross entropy loss with label smoothing regularizer.
Reference:
Szegedy et al. Rethinking the Inception Architecture for Computer Vision. CVPR 2016.
Equation: y = (1 - epsilon) * y + epsilon / K.
Args:
num_classes (int): number of classes.
epsilon (float): weight.
"""
def __init__(self, num_classes, epsilon=0.1, use_gpu=True):
super(CrossEntropyLabelSmooth, self).__init__()
self.num_classes = num_classes
self.epsilon = epsilon
self.use_gpu = use_gpu
self.logsoftmax = nn.LogSoftmax(dim=1)
def forward(self, inputs, targets):
"""
Args:
inputs: prediction matrix (before softmax) with shape (batch_size, num_classes)
targets: ground truth labels with shape (num_classes)
"""
log_probs = self.logsoftmax(inputs)
targets = torch.zeros(log_probs.size()).scatter_(1, targets.unsqueeze(1).data.cpu(), 1)
if self.use_gpu: targets = targets.npu()
targets = (1 - self.epsilon) * targets + self.epsilon / self.num_classes
loss = (- targets * log_probs).mean(0).sum()
return loss