Welcome to perte

A fast way to build lots of loss functions (fonction de perte in French) in Deep Learning.

Features

• Currently support:
  • Triplet loss and its variance
  • Angular margin penalty losses
  • Recall loss

Installation

To install with pip, use: pip install perte. If you install with pip, you should install PyTorch first by following the PyTorch installation instructions.

Usages

Simple Triplet loss

import torch
from perte import TripletLoss

## Initialize loss function
loss_fnc = TripletLoss(
    alpha=0.5, 
    reduction="mean", 
    device=torch.device("cpu")
)

## Compute the loss value
anchor_embd = torch.randn(1, 10)    ## features' dim = 10
positive_embd = torch.randn(1, 10)    ## features' dim = 10
negative_embd = torch.randn(1, 10)    ## features' dim = 10
loss_value = loss_fnc(anchor_embd, positive_embd, negative_embd)

Hard Mining Triplet loss

import torch
from perte import OnlineTripletLoss
from perte import AllTripletSelector
from perte import BatchHardTripletSelector
from perte import HardestNegativeTripletSelector
from perte import RandomNegativeTripletSelector
from perte import SemihardNegativeTripletSelector

## Initialize loss function
loss_fnc = OnlineTripletLoss(
    triplet_selector=AllTripletSelector,
    margin=0.5,
    reduction="mean", 
    device=torch.device("cpu")
)

## Compute the loss value
anchor_embd = torch.randn(1, 10)    ## features' dim = 10
positive_embd = torch.randn(1, 10)    ## features' dim = 10
negative_embd = torch.randn(1, 10)    ## features' dim = 10
loss_value = loss_fnc(anchor_embd, positive_embd, negative_embd)